AI-Based Smart Governance and Compliance Monitoring System for Coal Mines

1. Executive Snapshot

This research and technical dossier provides a production-grade blueprint for developing a centralized, offline-capable, AI-enabled smart governance and compliance monitoring platform for Coal India Limited (CIL) [PS]. The Indian coal mining ecosystem operates across eight major subsidiaries, hundreds of opencast and underground mine blocks, and thousands of field workers and contractors [KNOWN]. Currently, statutory compliance (under the Mines Act 1952 and Coal Mines Regulations 2017), safety inspection logs, environmental monitoring, and worker attendance are fragmented across legacy software, paper registers (e.g., DGMS Form 14/15), and isolated spreadsheets [PS, RESEARCHED]. This fragmented approach causes reporting delays, compliance gaps, lack of field accountability, and delayed administrative decision-making [PS]. The proposed system, MineGov-AI, bridges these operational gaps through a unified geo-spatial e-governance web platform paired with an offline-first mobile inspection app [INFERENCE]. It incorporates computer vision/OCR for document digitization, an XGBoost hazard prediction engine for proactive risk mitigation, automated multi-tier escalation workflows, and a SHA-256 cryptographic audit trail to enforce tamper-proof paperless governance across CIL mine sites [INFERENCE].

Fundamental System Objective: To establish a centralized, offline-resilient, AI-powered e-governance platform that unifies statutory compliance tracking, geo-tagged field inspections, predictive risk analytics, contractor management, and paperless workflow automation across Coal India Limited's mining ecosystem [INFERENCE].

2. Problem Deconstruction

The problem statement has been deconstructed into explicit requirements, preserves the original scope, and identifies key technical implications:

• REQ-A: Statutory Compliance Tracking
  • Source Text: "Digitally track statutory compliance requirements related to safety, environment, production, and labour regulations." [PS]
  • Interpretation: The platform must maintain a dynamic statutory matrix covering DGMS safety regulations (CMR 2017), CPCB/SPCB environmental consents (CTO/CTE), production quotas, and labour laws (Mines Act 1952, OSH&WC Code 2020) [RESEARCHED].
  • Implementation Implication: Database schema must support recurring compliance schedules, document attachments, approval hierarchies, and auto-generated statutory forms (e.g., Form 14, Form 15, Form XIII) [INFERENCE].
  • Tag: [PS]
• REQ-B: Real-time Inspection & Violation Monitoring
  • Source Text: "Enable real-time monitoring of inspections, observations, violations, and corrective actions." [PS]
  • Interpretation: Inspections logged in the field must instantly reflect on regional and corporate dashboards, with tracking of assigned Corrective and Preventive Actions (CAPA) until closure [INFERENCE].
  • Implementation Implication: Real-time websocket feed on web dashboard; state machine workflow for violation lifecycle (Open -> Under Review -> Corrective Action In Progress -> Verified & Closed) [INFERENCE].
  • Tag: [PS]
• REQ-C: Predictive AI & Risk Analytics
  • Source Text: "Use AI/analytics to identify high-risk areas, recurring compliance failures, and operational anomalies." [PS]
  • Interpretation: Machine learning models must process historical inspection logs, environmental sensor feeds, and shift reports to predict high-risk pit zones and flag recurring safety/environmental anomalies [INFERENCE].
  • Implementation Implication: Machine learning service running XGBoost for risk scoring per mine sector and Isolation Forest for anomaly detection in sensor/attendance data [INFERENCE].
  • Tag: [PS]
• REQ-D: Geo-tagged & Time-stamped Field Reporting Mobile App
  • Source Text: "Provide geo-tagged and time-stamped field reporting through mobile applications... with offline support." [PS]
  • Interpretation: Field safety officers must capture observations, photos, and worker attendance directly inside deep mine pits or remote blocks without cellular connectivity [INFERENCE].
  • Implementation Implication: Cross-platform Flutter mobile app with SQLite/WatermelonDB local storage, background vector-clock sync engine, EXIF metadata extraction, and GPS geofencing validation [INFERENCE].
  • Tag: [PS]
• REQ-E: Multi-Role Governance Dashboards
  • Source Text: "Integrate dashboards for mine officials, corporate management, and regulatory authorities." [PS]
  • Interpretation: Granular, role-based access control (RBAC) web dashboards tailored for Mine Managers (Colliery level), Subsidiary HQ Officers (e.g., BCCL, CCL, MCL HQ), and External Regulators (DGMS, CPCB) [RESEARCHED].
  • Implementation Implication: Web frontend built in React/Next.js with strict JWT/RBAC claims and specialized views (operational pit view, executive KPI view, regulatory compliance view) [INFERENCE].
  • Tag: [PS]
• REQ-F: Automated Workflows & Escalation Mechanisms
  • Source Text: "Generate automated alerts, reminders, compliance reports, and escalation mechanisms." [PS]
  • Interpretation: If an unmitigated safety violation exceeds its resolution SLA (e.g., 48 hours for gas accumulation or slope instability), the system must automatically escalate the ticket to higher corporate hierarchy [INFERENCE].
  • Implementation Implication: Asynchronous background job queue (Celery/Redis) with configurable SLA timers, push notifications, email/SMS gateway integration, and automated PDF report generation [INFERENCE].
  • Tag: [PS]
• REQ-G: Paperless Governance & OCR Digitization
  • Source Text: "Minimize manual paperwork... OCR/document digitization, workflow automation, blockchain-based audit trails..." [PS]
  • Interpretation: Legacy physical registers, paper inspection logs, and handwritten statutory forms must be scanned and automatically converted into structured digital records with immutable verification [INFERENCE].
  • Implementation Implication: OCR pipeline (Tesseract/OpenCV) for extracting text from scanned forms, combined with SHA-256 cryptographic hash chaining for tamper-evident audit logs [INFERENCE].
  • Tag: [PS]
• REQ-H: Scalability Across Subsidiaries
  • Source Text: "Be scalable for deployment across multiple mines and subsidiaries." [PS]
  • Interpretation: Architecture must support multi-tenant hierarchy handling Coal India's 8 main subsidiaries (BCCL, CCL, ECL, MCL, NCL, SECL, WCL, CMPDI) and hundreds of individual mining blocks [RESEARCHED].
  • Implementation Implication: Multi-tenant database design with tenant-id scoping, stateless API microservices, containerized Docker deployment, and PostGIS spatial indexing [INFERENCE].
  • Tag: [PS]

3. Stakeholders & User Needs

4. Requirements Traceability

5. Constraints & Environment Realities

• Geography & Harsh Field Conditions: Coal India mines are located across remote industrial belts (e.g., Dhanbad, Korba, Singrauli, Asansol) [RESEARCHED]. Opencast pits experience severe dust, high mud during monsoons, and extreme temperatures (up to 48°C), while underground mines suffer from extreme humidity, continuous airborne coal dust, and hazardous explosive atmospheric conditions (Methane/CO) [KNOWN].
• Connectivity Realities: Mobile cellular reception drops to zero inside deep opencast pits (>100 meters deep) and underground galleries [KNOWN]. Cloud connectivity cannot be assumed during field inspections. All field applications must be fully functional offline for at least 72 hours [INFERENCE].
• Hardware & Explosive Atmosphere Constraints: Standard commercial smartphones are not flameproof. In underground mines (DGMS Degree I, II, III gaseous mines), devices must strictly comply with IS/IEC 60079 intrinsically safe standards or be operated only in designated safe zones (e.g., surface office / pithead) [KNOWN, RESEARCHED]. Field smartphones used in opencast mines require IP68 water/dust sealing and drop-resistant casing [INFERENCE].
• Power Supply Instability: Mining field offices frequently experience power fluctuations and extended outages [KNOWN]. Field inspection devices must operate on optimized battery power (minimal background GPS polling), and edge gateways must include UPS power backup [INFERENCE].
• Language & Digital Literacy: Field supervisors and contractor staff possess varying levels of digital literacy [KNOWN]. Interfaces must be minimalist, visually intuitive (color-coded status badges, icon-heavy), and support multilingual input (English, Hindi, Bengali, Odia) with local voice-to-text dictation capabilities [INFERENCE].

6. Research Questions

1. What specific statutory forms under the Coal Mines Regulations (CMR) 2017 and Mines Rules 1955 must be digitized to satisfy DGMS inspection standards?
   • Verification: CMR 2017 Regulation 104 (Safety Management Plans), Form 14 (Accident/Incident Register), Form 15 (Notice of Dangerous Occurrence), and Mines Rules Form XIII (Contractor Labour Register) [RESEARCHED].
2. What are the exact integration boundaries with Coal India's existing Enterprise Resource Planning (ERP) systems?
   • Verification: CIL uses SAP ERP (Project PRIDE) for financial and HR data, alongside CMSMS/Khanan Prahari for illegal mining surveillance, and CIL ICIS for contractor tracking [RESEARCHED]. The solution must interface via secure REST/OData endpoints [INFERENCE].
3. How can mobile application integrity and GPS location spoofing be prevented in deep pit reporting?
   • Verification: Implement hardware-backed GPS location verification, cell tower triangulation fallback, OS mock-location detection flags, and EXIF metadata extraction [INFERENCE].
4. What offline data synchronization strategy best prevents data loss and resolves concurrent edit conflicts?
   • Verification: Utilize timestamped vector clocks combined with server-authoritative conflict resolution rules (regulatory status state machine overrides local state) [INFERENCE].
5. Is blockchain strictly required for audit trails, or can a lighter cryptographic alternative suffice?
   • Verification: A SHA-256 hash-chained append-only database log provides equivalent tamper-evidence with significantly lower computational overhead and zero infrastructure complexity for a hackathon MVP [INFERENCE].

7. Data Requirements & Sources

8. Existing Solutions & Gap Analysis

• 1. Coal India ODRS (Real-Time Contravention Reporting System)
  • Description: CIL's internal portal for reporting unsafe acts, unsafe conditions, and near-misses via QR code scanning [RESEARCHED].
  • Strengths: Quick reporting interface for surface staff; centralized reporting [RESEARCHED].
  • Weaknesses: Limited offline functionality in deep pits; lacks predictive risk analytics; no integrated document OCR or contractor compliance tracking [INFERENCE].
  • Relevance: Direct official baseline for safety reporting in CIL [RESEARCHED].
  • Gap: Lacks predictive ML, offline field sync, integrated GIS spatial heatmaps, and automated statutory form generation [INFERENCE].
  • Source: Official CIL Portal (https://www.coalindiaodrs.live/) [RESEARCHED].
  • Confidence: [RESEARCHED]
• 2. CMSMS & Khanan Prahari App (Ministry of Coal / CMPDI / BISAG-N)
  • Description: Satellite-based surveillance system to detect illegal coal mining within 100 meters of coalfield boundaries [RESEARCHED].
  • Strengths: Strong geo-spatial mapping; crowdsourced citizen incident reporting [RESEARCHED].
  • Weaknesses: Focused exclusively on external illegal mining deterrence; does not handle internal mine safety, statutory DGMS compliance, or contractor management [INFERENCE].
  • Relevance: Demonstrates Ministry of Coal's commitment to GIS spatial monitoring [RESEARCHED].
  • Gap: Does not address internal operational governance, mine safety inspections, or environmental CAPA tracking [INFERENCE].
  • Source: Ministry of Coal / CMPDI Official Documentation (https://www.cmpdi.co.in/) [RESEARCHED].
  • Confidence: [RESEARCHED]
• 3. Commercial DGMS Mining Software (e.g., iCeipts DGMS Platform)
  • Description: Commercial SaaS product designed for Indian mining companies to replace paper registers with automated Form 14/15 reporting [RESEARCHED].
  • Strengths: Comprehensive mapping to Indian statutory mining laws; automated register generation [RESEARCHED].
  • Weaknesses: Proprietary commercial software; lacks deep enterprise integration with CIL's SAP ERP; high licensing cost; missing native AI-driven hazard prediction engines [INFERENCE].
  • Relevance: Proves market demand for automated DGMS compliance software [RESEARCHED].
  • Gap: Lacks native multi-subsidiary governance hierarchy, open-source AI risk modeling, and cryptographic tamper-evident audit trails [INFERENCE].
  • Source: Industry documentation (https://iceipts.com/products/dgms-compliance) [RESEARCHED].
  • Confidence: [RESEARCHED]

9. Candidate Solution Approaches

+-----------------------------------------------------------------------------------------------------------------------------------+
| CANDIDATE SOLUTION COMPARISON MATRIX                                                                                              |
+------------------------------------+------------------------------------+------------------------------------+--------------------+
| Dimension                          | Approach 1: Digitized Web Portal   | Approach 2: Cloud-Native DL/LLM    | Approach 3: Hybrid |
|                                    | (Rule-Based Monolith)              | Microservices                      | GIS AI (Proposed)  |
+------------------------------------+------------------------------------+------------------------------------+--------------------+
| Mechanism                          | Standard CRUD web portal with      | Heavy deep learning, vision models,| Offline-first      |
|                                    | basic relational forms and static  | LLM regulatory chatbots, and cloud | Flutter app, GIS   |
|                                    | threshold notifications.           | microservices architecture.        | PostGIS backend,   |
|                                    |                                    |                                    | XGBoost ML, OCR.   |
| Problem Fit                        | Low - Fails on offline pit         | Medium - High cloud dependency;    | High - Tailored for|
|                                    | connectivity and predictive risk.  | impractical inside deep pits.      | field realities and|
|                                    |                                    |                                    | CIL multi-tier.    |
| Accuracy / Analytics               | Minimal - Deterministic rules only.| High - But prone to LLM hallucina- | High - Explainable |
|                                    | No ML/predictive capacity.         | tions in statutory compliance.     | ML risk scoring &  |
|                                    |                                    |                                    | exact rule engine. |
| Data Requirements                  | Low - Basic SQL tables.            | High - Requires massive training   | Moderate - Structured|
|                                    |                                    | datasets for LLM and Vision models.| logs, CSVs, GIS.   |
| Complexity                         | Low - Monolithic web stack.        | Extremely High - Complex cloud     | Moderate - Balanced|
|                                    |                                    | pipelines & high infrastructure.   | modular services.  |
| Explainability                     | High - Pure deterministic rules.   | Low - Black-box deep learning and  | High - Tree-based  |
|                                    |                                    | generative LLMs.                   | ML feature weights.|
| Hackathon Feasibility              | High - Very easy to build.         | Low - Too many moving cloud parts; | High - Highly      |
|                                    |                                    | hard to complete in hackathon.     | buildable MVP.     |
| Production Feasibility             | Low - Fails CIL enterprise scale   | Low - Prohibitive cloud cost and   | High - Scalable,   |
|                                    | and offline field requirements.    | offline failure points.            | cost-effective.    |
+------------------------------------+------------------------------------+------------------------------------+--------------------+

10. Recommended Approach & Why

Selected Approach: Approach 3 — Hybrid Geo-Spatial AI Platform with Offline Mobile Engine, Automated Workflows, and Cryptographic Audit Trail.

• Why Alternatives Were Rejected:
  • Approach 1 (Rule-Based Monolith) was rejected because it fails to satisfy REQ-C (Predictive AI/Analytics for high-risk identification) and REQ-D (Offline geo-tagged mobile operations) [INFERENCE]. A simple web CRUD app cannot handle zero-connectivity deep pit inspections or proactively flag recurring violation patterns [INFERENCE].
  • Approach 2 (Pure Cloud-Native DL/LLM Microservices) was rejected due to unrealistic deployment constraints in mining environments [INFERENCE]. Generative LLMs and heavy cloud computer vision models cannot run offline inside deep pits, present high cloud latency/costs, and introduce unacceptable hallucination risks when evaluating statutory safety compliance [INFERENCE].
• Why Approach 3 Wins:
  1. Field Reality Alignment: Uses a lightweight Flutter mobile client with SQLite/WatermelonDB, allowing field inspectors to capture observations, photo evidence, and worker attendance offline, syncing automatically when reaching surface Wi-Fi or cellular range [INFERENCE].
  2. Explainable AI: Uses tree-based gradient boosting (XGBoost) for hazard risk scoring and Isolation Forest for operational anomaly detection [INFERENCE]. Feature importance weights (e.g., historical pit slope instability, rainfall intensity, missed maintenance cycles) are fully transparent and explainable to DGMS safety auditors [INFERENCE].
  3. Audit Integrity: Replaces complex, resource-heavy public blockchains with a highly efficient, deterministic SHA-256 cryptographic hash-chain ledger, ensuring that compliance logs and closure approvals cannot be retroactively altered [INFERENCE].
  4. Hackathon Feasibility: Can be fully prototyped by a 4-6 student team within the build window using open-source tools (Flutter, Node.js/Python FastAPI, PostgreSQL/PostGIS, React, Scikit-Learn/XGBoost) [INFERENCE].

11. Technical Architecture

+---------------------------------------------------------------------------------------------------+
| SYSTEM ARCHITECTURE DIAGRAM: MINEGOV-AI PLATFORM                                                  |
+---------------------------------------------------------------------------------------------------+

[ PRESENTATION LAYER ]
  +---------------------------+  +---------------------------+  +---------------------------+
  | Mine Officer Web App      |  | Subsidiary HQ Executive   |  | Regulatory Portal (DGMS)  |
  | (React / Next.js + Mapbox)|  | Dashboard (React + Recharts) | (Read-Only + Audit View) |
  +---------------------------+  +---------------------------+  +---------------------------+
               ^                              ^                              ^
               +------------------------------+------------------------------+
                                              | (REST API / HTTPS / WSS)
[ API GATEWAY & SECURITY LAYER ]
  +-------------------------------------------------------------------------------------------------+
  | FastAPI / Express API Gateway | NGINX Reverse Proxy | OAuth2 / JWT Auth | Role-Based Access Control|
  +-------------------------------------------------------------------------------------------------+
                                              |
[ APPLICATION & WORKFLOW ENGINE LAYER ]
  +--------------------+  +--------------------+  +--------------------+  +--------------------+
  | Inspection & CAPA  |  | Dynamic Statutory  |  | Automated SLA &    |  | Contractor &       |
  | Management Service |  | Compliance Service |  | Escalation Engine  |  | Manpower Service   |
  +--------------------+  +--------------------+  +--------------------+  +--------------------+
                                              |
[ AI / ANALYTICS & DIGITIZATION ENGINE ]
  +------------------------------------+  +------------------------------------+  +-----------------+
  | Predictive Risk Module             |  | Anomaly Detection Engine           |  | OCR Pipeline    |
  | (XGBoost / LightGBM Hazard Scoring)|  | (Isolation Forest Sensor/Logs)     |  | (OpenCV + Tesseract)
  +------------------------------------+  +------------------------------------+  +-----------------+
                                              |
[ STORAGE & AUDIT LAYER ]
  +------------------------------------+  +------------------------------------+  +-----------------+
  | PostGIS Spatial Database           |  | Redis Cache & Queue                |  | SHA-256 Crypto  |
  | (Mines, Violations, Workflows)     |  | (Background SLA Jobs / WebSockets) |  | Audit Ledger    |
  +------------------------------------+  +------------------------------------+  +-----------------+
                                              ^
                                              | (Delta Sync Engine / JSON Batches)
[ EDGE / MOBILE LAYER (OFFLINE FIELD OPERATION) ]
  +-------------------------------------------------------------------------------------------------+
  | Flutter Cross-Platform Mobile App (Android / iOS)                                              |
  |  - Encrypted SQLite Local Storage (SQLCipher)                                                   |
  |  - Vector Clock Delta Sync Engine & Background Upload Queue                                     |
  |  - EXIF Photo Metadata & GPS Geofencing Module                                                 |
  |  - Multilingual Offline Voice-to-Text Input (Vosk / Whisper Lite)                               |
  +-------------------------------------------------------------------------------------------------+

Detailed Data-Flow Narrative

1. Field Observation Capture (Offline State): A Mine Safety Officer in an underground gallery logs an unsafe roof strata condition using the Flutter mobile app [INFERENCE]. The app captures photo evidence, records GPS coordinates (or gallery node ID), logs exact timestamp, and saves the encrypted record locally into SQLite [INFERENCE].
2. Edge Sync Trigger: Upon returning to the surface office, the app detects an active network connection and initiates delta synchronization via REST endpoints [INFERENCE]. The vector clock engine compares local and server timestamps, resolves non-conflicting records, and pushes the payload [INFERENCE].
3. Ingestion & Cryptographic Hashing: The API Gateway validates the JWT token, extracts user role claims, and forwards the record to the Inspection Service [INFERENCE]. The service calculates a SHA-256 hash incorporating the previous log's hash, record contents, and inspector ID, appending it to the cryptographic audit ledger [INFERENCE].
4. Risk Scoring & Anomaly Evaluation: The record triggers the AI Engine [INFERENCE]. XGBoost evaluates current pit parameters (roof condition, local humidity, historic incidents) and outputs a Hazard Risk Score (0-100) [INFERENCE]. If the score exceeds 75, the anomaly engine flags the sector as a "High-Risk Zone" [INFERENCE].
5. Automated Escalation & Real-Time Notification: An unmitigated critical violation triggers the SLA Engine [INFERENCE]. Redis queues a background task with a 24-hour countdown timer [INFERENCE]. If the Colliery Agent fails to record a CAPA action within 24 hours, the system automatically escalates the ticket via SMS/Email to the Subsidiary General Manager (Safety) and updates the web dashboard via WebSocket push [INFERENCE].
6. Regulatory Access & Statutory Reporting: DGMS inspectors log into the read-only Regulatory Portal to inspect the digital Form 14 register [INFERENCE]. The system dynamically renders the verified inspection records with cryptographic hash proofs, allowing one-click PDF export [INFERENCE].

12. Technology Stack

13. AI/ML Justification

1. Hazard Risk Predictor

• Purpose: Predict high-risk mine sectors and likelihood of safety violations in upcoming shifts [INFERENCE].
• Model Type: XGBoost Classifier / Regressor [INFERENCE].
• Input: Historical sector violation logs, shift length, rainfall (mm), equipment age, recent maintenance flags, geological strata stability index [INFERENCE].
• Output: Pit Sector Risk Index (0.0 to 1.0) and top 3 contributing risk factors [INFERENCE].
• Data Required: Tabular dataset of 2,000+ past inspection logs with historical environmental and operational parameters [ASSUMPTION].
• Why AI is Preferable: Complex non-linear interactions between environmental variables (e.g., rainfall + slope angle + blast frequency) cannot be captured by simple static threshold rules [INFERENCE].
• Evaluation Metrics: ROC-AUC (>0.85), Precision, and Recall on high-risk hazard classes [INFERENCE].
• Failure Mode Handling: If model inference fails, default to conservative rule-based maximum safety classification [INFERENCE].
• Human Oversight: Risk scores serve as advisory highlights on dashboard; humans perform all physical site closures [INFERENCE].

2. Operational Anomaly & Near-Miss Detector

• Purpose: Identify abnormal environmental sensor readings (AQMS/gas) or fraudulent attendance patterns [INFERENCE].
• Model Type: Isolation Forest (Unsupervised Anomaly Detection) [INFERENCE].
• Input: Continuous stream of PM10, PM2.5, Methane, CO sensor values, and worker check-in timestamps [INFERENCE].
• Output: Anomaly Score (-1 for outlier, +1 for normal) with flagged outlier parameters [INFERENCE].
• Data Required: Baseline time-series sensor telemetry [ASSUMPTION].
• Why AI is Preferable: Detects subtle multivariate drifts before fixed hard thresholds are breached [INFERENCE].
• Evaluation Metrics: False Positive Rate (<5%), Precision at K [INFERENCE].
• Failure Mode Handling: Fall back to strict static threshold alerts (e.g., Methane > 0.75% triggers immediate hard alert) [INFERENCE].
• Human Oversight: Safety officer must physically inspect and confirm flagged sensor anomalies [INFERENCE].

3. Statutory Document Digitizer (OCR Engine)

• Purpose: Extract structured text fields from scanned physical Form 14/15 and paper inspection sheets [PS, INFERENCE].
• Model Type: Tesseract OCR with OpenCV image preprocessing pipeline (deskew, binarization, noise removal) [INFERENCE].
• Input: Scanned JPG/PNG/PDF image of statutory form [INFERENCE].
• Output: Extracted JSON key-value pairs (Inspector Name, Date, Violation Category, Location Code) [INFERENCE].
• Data Required: 20+ sample form templates [ASSUMPTION].
• Why AI is Preferable: Automates manual data entry from physical paper registers [INFERENCE].
• Evaluation Metrics: Character Error Rate (CER < 5%), Field Extraction Accuracy (>90%) [INFERENCE].
• Failure Mode Handling: Highlight low-confidence text fields in red on the web interface for human manual correction [INFERENCE].
• Human Oversight: Mandatory human approval step before committing OCR-extracted data into official statutory database [INFERENCE].

Where AI Must NOT Be Used

AI must NOT be used to automatically issue statutory penalties, grant mine safety clearances, override safety officers' observations, or automatically close compliance tickets [INFERENCE]. All regulatory actions must remain strictly deterministic and enforce mandatory human approval to prevent legal liability and algorithmic bias [INFERENCE].

14. Data Architecture & Offline/Low-Network Design

Database Schema (Core Relational Entities)

-- Mine Subsidiaries and Sites
CREATE TABLE mine_sites (
    mine_id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
    subsidiary_code VARCHAR(10) NOT NULL, -- e.g., 'BCCL', 'CCL', 'MCL'
    mine_name VARCHAR(100) NOT NULL,
    boundary_geom GEOMETRY(Polygon, 4326),
    created_at TIMESTAMP WITH TIME ZONE DEFAULT CURRENT_TIMESTAMP
);

-- Field Inspection Logs
CREATE TABLE inspections (
    inspection_id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
    mine_id UUID REFERENCES mine_sites(mine_id),
    inspector_id UUID NOT NULL,
    sector_name VARCHAR(50) NOT NULL,
    location_geom GEOMETRY(Point, 4326),
    observation_text TEXT NOT NULL,
    hazard_level VARCHAR(20) CHECK (hazard_level IN ('CRITICAL', 'HIGH', 'MEDIUM', 'LOW')),
    photo_urls TEXT[],
    sync_status VARCHAR(20) DEFAULT 'SYNCED',
    created_at TIMESTAMP WITH TIME ZONE NOT NULL,
    client_timestamp TIMESTAMP WITH TIME ZONE NOT NULL
);

-- Corrective and Preventive Actions (CAPA)
CREATE TABLE capa_tickets (
    ticket_id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
    inspection_id UUID REFERENCES inspections(inspection_id),
    assigned_official_id UUID NOT NULL,
    action_required TEXT NOT NULL,
    status VARCHAR(20) DEFAULT 'OPEN' CHECK (status IN ('OPEN', 'IN_PROGRESS', 'RESOLVED', 'CLOSED')),
    sla_due_at TIMESTAMP WITH TIME ZONE NOT NULL,
    escalation_level INT DEFAULT 0,
    resolved_at TIMESTAMP WITH TIME ZONE
);

-- Cryptographic Audit Ledger
CREATE TABLE audit_ledger (
    entry_id BIGSERIAL PRIMARY KEY,
    entity_type VARCHAR(50) NOT NULL,
    entity_id UUID NOT NULL,
    action VARCHAR(20) NOT NULL,
    payload_json JSONB NOT NULL,
    previous_hash VARCHAR(64) NOT NULL,
    current_hash VARCHAR(64) NOT NULL,
    recorded_at TIMESTAMP WITH TIME ZONE DEFAULT CURRENT_TIMESTAMP
);

Offline Synchronization & Bandwidth Reduction Strategies

• Local Storage & Encryption: Mobile device uses SQLite encrypted with SQLCipher (256-bit AES) [INFERENCE]. Unsynced records are flagged with local status PENDING_SYNC [INFERENCE].
• Vector Clock Delta Sync: Each inspection record maintains a vector timestamp (client_timestamp, server_timestamp) [INFERENCE]. Upon connection re-establishment, the app sends only changed delta records since the last successful sync checkpoint [INFERENCE].
• Conflict Handling:
  1. Regulatory & Safety Status: Server-Authoritative (if an admin closed a ticket on the server, local mobile device state is updated) [INFERENCE].
  2. Field Inspection Logs: Client-Append (new field notes logged offline are never overwritten by server state; they are appended with a unique UUID) [INFERENCE].
• Bandwidth Reduction:
  • Image Compression: Photos captured in the field are downsampled on-device to WebP format (80% quality reduction, max 1080p resolution), reducing payload size from ~5MB to ~250KB per photo [INFERENCE].
  • Payload Compression: Sync payloads are compressed using Gzip before transfer over HTTP/2 [INFERENCE].
  • Offline UX: Visual badge displays connectivity status (Green = Online, Amber = Offline - Queued Records). Inspector can continue taking logs seamlessly without blocking UI dialogs [INFERENCE].

15. Security, Privacy & Governance

• Personal & Location Data Protection: Worker biometric/attendance logs and inspector real-time GPS locations are classified as Sensitive Personal Data under India's Digital Personal Data Protection (DPDP) Act 2023 [RESEARCHED]. Precise GPS tracking is active only when an inspector explicitly initiates a field inspection report [INFERENCE].
• Authentication & Access Control:
  • OAuth2 with JWT (JSON Web Tokens) enforcing strict Role-Based Access Control (RBAC) [INFERENCE].
  • Roles: FIELD_INSPECTOR, MINE_MANAGER, SUBSIDIARY_HQ_ADMIN, REGULATORY_AUDITOR [INFERENCE].
  • Multi-Factor Authentication (MFA) required for high-privilege corporate and regulatory logins [INFERENCE].
• Encryption Standards: AES-256 encryption at rest for database tables and local mobile SQLite files; TLS 1.3 protocol for all data in transit [INFERENCE].
• Tamper-Evident SHA-256 Audit Ledger: Every statutory submission, violation edit, or approval triggers a SHA-256 cryptographic hash computation: $$\text{Current Hash} = \text{SHA-256}(\text{Previous Hash} + \text{Payload JSON} + \text{User ID} + \text{Timestamp})$$ Any attempt to modify past statutory records in the database invalidates the hash chain, immediately alerting system auditors [INFERENCE].
• Regulatory & Legal Safeguard: The platform explicitly functions as an enterprise decision-support and workflow automation tool [INFERENCE]. Final legal responsibility for mine safety remains with the certified Mine Manager as mandated under Regulation 27 of the Coal Mines Regulations 2017 [RESEARCHED].

16. MVP Scope (Hackathon Build)

+---------------------------------------------------------------------------------------------------+
| MVP SCOPE (36-HOUR HACKATHON BUILD vs PRODUCTION)                                                |
+---------------------------------------------------------------------------------------------------+

                          [ MUST BUILD / IN SCOPE FOR HACKATHON ]
  +---------------------------------------------------------------------------------------+
  | 1. Geo-Tagged Mobile Inspection App (Flutter)                                         |
  |    - Offline inspection logging with WebP image capture & location tagging             |
  |    - Local SQLite storage with auto vector-clock background sync                       |
  |                                                                                       |
  | 2. Centralized Multi-Role Web Dashboard (React/Next.js)                               |
  |    - Mine Official View (CAPA ticket queue & SLA tracker)                             |
  |    - Subsidiary HQ View (Multi-mine risk heatmap & KPI metrics)                       |
  |    - Regulatory DGMS View (Form 14/15 digital viewer with PDF export)                 |
  |                                                                                       |
  | 3. AI Risk & Anomaly Engine (Python FastAPI)                                          |
  |    - XGBoost model evaluating pit hazard risk scores                                  |
  |    - Tesseract OCR pipeline parsing scanned Form 14 images                            |
  |                                                                                       |
  | 4. Core Workflow & Security Infrastructure                                             |
  |    - Redis SLA countdown timer triggering automated email alerts                      |
  |    - SHA-256 cryptographic audit ledger for tamper-proof logging                      |
  +---------------------------------------------------------------------------------------+

                     [ DESCRIBED ONLY / OUT OF SCOPE FOR HACKATHON ]
  +---------------------------------------------------------------------------------------+
  | 1. Direct Live Integration with CIL Production SAP ERP System (Simulated via Mock API) |
  | 2. Physical Drone LiDAR / High-Res Satellite Feed Processing (Simulated via GeoJSON)  |
  | 3. Hardware intrinsically safe certification testing (IS/IEC 60079)                   |
  | 4. Enterprise-wide Single Sign-On (SSO) with CIL Active Directory                      |
  +---------------------------------------------------------------------------------------+

17. Live Demo Simulation Plan

To present a compelling, technically honest live demonstration without relying on unavailable live government feeds or mining hardware, the team will execute the following simulation plan:

1. Simulated Field Sensor Telemetry:
   • Simulation: Run a background Python script generating continuous MQTT telemetry streams simulating AQMS sensors (PM10, Methane %, CO ppm) for 3 virtual mine pits [ASSUMPTION].
   • Demo Action: Dynamically inject a "Methane Spike (1.2%)" payload via script during the presentation, demonstrating real-time dashboard alert triggers [INFERENCE].
2. Offline Mobile Field Journey (Flight Mode Test):
   • Simulation: Presenter turns mobile phone (or Android Emulator) to Airplane Mode live on stage [INFERENCE].
   • Demo Action: Log a critical safety observation ("Unstable Highwall Slope") with a live photo capture [INFERENCE]. Show local SQLite saving. Turn Airplane Mode OFF; demonstrate automated background sync and instant WebSocket update on the live web dashboard [INFERENCE].
3. OCR Form Digitization Live Upload:
   • Simulation: Use a pre-prepared scanned image of a handwritten DGMS Form 14 register [ASSUMPTION].
   • Demo Action: Upload file via web UI. Show Tesseract OCR pipeline processing in <3 seconds, auto-filling form fields (Inspector Name, Violation Category, Location Code) on screen [INFERENCE].
4. SLA Escalation & Cryptographic Audit Verification:
   • Simulation: Set Redis SLA timer to 15 seconds for demo purposes [ASSUMPTION].
   • Demo Action: Allow a critical violation ticket to expire. Watch SLA trigger escalate ticket status to "Subsidiary GM View" with instant email notification. Click "Verify Audit Trail" to display SHA-256 hash validation [INFERENCE].

18. Implementation Roadmap

+---------------------------------------------------------------------------------------------------+
| IMPLEMENTATION ROADMAP (Target Deadline: 30 September 2026)                                       |
+---------------------------------------------------------------------------------------------------+

[Phase 1: Architecture & Data Modeling] (Weeks 1-3)
  ├── Finalize PostgreSQL/PostGIS schemas & API contracts
  └── Collect & synthesize DGMS compliance forms & 2-year mock inspection logs

[Phase 2: Core Mobile Development] (Weeks 4-7)
  ├── Build Flutter offline inspection client with SQLite/SQLCipher
  └── Implement vector-clock synchronization engine & EXIF photo capture

[Phase 3: Backend Services & GIS] (Weeks 8-11)
  ├── Develop Python FastAPI REST/WSS microservices
  ├── Set up PostGIS database & Mapbox spatial pit layer
  └── Construct Redis SLA queue & automated notification pipeline

[Phase 4: AI Engine & OCR Pipeline] (Weeks 12-14)
  ├── Train XGBoost Hazard Risk Model on tabular mining logs
  └── Build OpenCV + Tesseract OCR extraction service for statutory forms

[Phase 5: Web Dashboard & RBAC] (Weeks 15-17)
  ├── Develop Next.js multi-tenant web portals (Mine, HQ, Regulatory)
  └── Implement SHA-256 cryptographic audit ledger verification view

[Phase 6: Integration, Security & Simulation] (Weeks 18-20)
  ├── Perform offline sync stress testing & security hardening
  └── Build live demo telemetry simulators & field trial verification

  ===> FINAL SIH SUBMISSION & DEMO READY: 30 SEPTEMBER 2026 [PS] <===

19. Testing & Validation Plan

• Functional Testing: End-to-end verification of ticket lifecycles (Open -> In Progress -> Resolved -> Closed) across all 4 user roles [INFERENCE].
• Performance & Scale Testing: Load test FastAPI endpoints using Locust, targeting 500 concurrent sync requests with response latency < 200ms [INFERENCE].
• Usability Testing: Evaluate mobile UI with non-technical users to ensure complete inspection logging requires < 4 taps [INFERENCE].
• AI Model Evaluation:
  • XGBoost Hazard Risk Predictor evaluated on 20% holdout test set: Target ROC-AUC $\ge 0.85$, Precision $\ge 0.80$, Recall $\ge 0.82$ [INFERENCE].
  • OCR Pipeline evaluated on 20 test scans: Target field recognition accuracy $\ge 90%$ [INFERENCE].
• False Positive / False Negative Stress Testing: Test Anomaly Detector against normal operational fluctuations (e.g., dust spikes due to heavy blasting) to ensure false alert rate stays below 5% [INFERENCE].
• Offline Synchronization Testing: Simulate extreme network degradation (packet loss 50%, latency 3000ms, sudden disconnects) during active mobile sync to verify zero data loss or duplicate row insertions in SQLite/Postgres [INFERENCE].
• Security Testing: OWASP Top 10 vulnerability scan, SQL injection checks, JWT key leakage tests, and cryptographic hash chain tampered row detection verification [INFERENCE].

20. Risk Register

21. Cost & Scalability

Hackathon Build

• Infrastructure Cost: $0 (Free Tier Services) [INFERENCE].
  • Hosting: Vercel (Next.js Frontend) & Render/Koyeb (FastAPI Backend) Free Tiers [INFERENCE].
  • Database: Supabase / Neon PostgreSQL (Free Tier 500MB) [INFERENCE].
  • SMS/Email: Twilio / SendGrid trial accounts [INFERENCE].
• Hardware Cost: $0 (Utilizing team members' Android/iOS smartphones and laptops) [INFERENCE].

Production System (CIL Scale: ~300 Mines, 8 Subsidiaries)

• Cloud / On-Premise Infrastructure: $15,000 - $30,000 / month [INFERENCE].
  • High-availability Kubernetes Cluster (EKS / On-Premise CIL Data Center) for API microservices [INFERENCE].
  • PostgreSQL Enterprise Cluster (Primary-Replica) with PostGIS spatial extensions [INFERENCE].
  • Dedicated Redis Enterprise Cluster for multi-subsidiary SLA queues [INFERENCE].
  • Self-hosted GeoServer tile servers for GIS pit mapping [INFERENCE].
• Field Hardware: Ruggedized IP68 Android devices for opencast field officers (~$300 / unit x 2,000 officers = $600,000 initial capital expense) [INFERENCE].
• Scalability Architecture: Horizontal pod autoscaling (HPA) on microservices; database connection pooling (PgBouncer); CDN static asset caching; PostGIS spatial partitioning by subsidiary ID [INFERENCE].

22. Differentiation

+---------------------------------------------------------------------------------------------------+
| COMPETITIVE DIFFERENTIATION MATRIX                                                                |
+------------------------------------+----------------------------------+---------------------------+
| Feature / Dimension                | Generic Hackathon Submissions    | MineGov-AI Platform       |
+------------------------------------+----------------------------------+---------------------------+
| Offline Operation                  | Basic online web forms; breaks   | True offline-first SQLite |
|                                    | in deep pits without cloud.      | with vector-clock sync.   |
| AI Implementation                  | Generic LLM wrapper chatbot      | Explainable XGBoost       |
|                                    | prone to hallucinations.         | risk scoring & OCR.       |
| Audit Security                     | Standard mutable SQL table;      | Immutable SHA-256         |
|                                    | easily tampered post-incident.   | cryptographic hash chain. |
| Statutory Alignment                | Generic safety checklist.        | Built for DGMS CMR 2017 & |
|                                    |                                    | Form 14/15/XIII workflows.|
| Enterprise Hierarchy               | Single-user / single-mine view.  | Multi-tenant: Colliery,   |
|                                    |                                    | HQ, and DGMS roles.       |
+------------------------------------+----------------------------------+---------------------------+

23. Future Scope

1. Satellite Synthetic Aperture Radar (SAR) Mine Subsidence Monitoring: Integrate Sentinel-1 SAR interferometry data to detect millimeter-level ground subsidence and slope instability in opencast highwalls automatically [RESEARCHED, INFERENCE].
2. Underground Miner Wearable IoT Integration: Connect smart wearable gas detectors and heart-rate bands via LoRaWAN mesh networks in underground galleries to stream personal safety telemetry to surface dashboards [INFERENCE].
3. Autonomous Drone Volumetric Stockpile Analysis: Ingest drone survey point cloud data to measure coal stockpile volumes accurately and cross-verify production logs automatically [RESEARCHED, INFERENCE].
4. Generative Regulatory Copilot (Local Fine-Tuned Model): Deploy a fully offline, locally hosted Llama-3-8B model fine-tuned on the Mines Act 1952 and CMR 2017 to assist mine managers in instantly drafting legally compliant CAPA responses [INFERENCE].

24. Assumptions & Open Items Log

Consolidated Assumptions & Validation Log

• [ASSUMPTION] Mock spatial boundaries for CIL opencast mines (e.g., Gevra OCP) accurately reflect real-world pit geometries for GeoJSON demonstration [Section 7].
• [ASSUMPTION] Synthetic inspection logs across 5 mine blocks accurately mimic historical distribution of DGMS safety violations [Section 7].
• [ASSUMPTION] 20 sample scanned images of Form 14/15 are sufficient to demonstrate OCR field extraction feasibility during the hackathon [Section 7].
• [ASSUMPTION] Simulated MQTT streams provide a technically valid proxy for real-time CIL CAAQMS sensor telemetry [Section 17].
• [NEEDS VALIDATION] Live API access specifications, authentication tokens, and OData schemas for Coal India's production SAP ERP system (Project PRIDE) [Section 7].
• [NEEDS VALIDATION] Specific DGMS circular guidelines regarding digital tablet usage policies inside underground Degree II and Degree III flammable gas mines [Section 5].

Final Recommendation

Build MineGov-AI using Approach 3 (Hybrid Geo-Spatial AI Platform with Offline Mobile Engine). This architecture addresses the field realities of Indian coal mining (zero pit connectivity, harsh environments, multi-tier organizational hierarchy) while satisfying all explicit requirements of SIH26024 [PS, INFERENCE].

Recommended Stack

• Mobile: Flutter + SQLite (SQLCipher) + Vector Clock Sync Engine [INFERENCE].
• Web Dashboard: React.js / Next.js + Mapbox GL / Leaflet [INFERENCE].
• Backend: Python FastAPI + Redis / Celery [INFERENCE].
• Database: PostgreSQL + PostGIS (Spatial) [INFERENCE].
• AI/ML & OCR: XGBoost + Isolation Forest + OpenCV + Tesseract OCR [INFERENCE].
• Audit Layer: Cryptographic SHA-256 Hash-Chain [INFERENCE].

Biggest Dependency

The reliability of local SQLite device storage and background synchronization protocols when inspectors transition between zero-connectivity pits and network-connected surface offices [INFERENCE].

Biggest Risk

User resistance from traditional field staff accustomed to physical paper registers, and hardware usage restrictions inside underground gaseous mine zones [KNOWN, INFERENCE].

Biggest Technical Challenge

Implementing a rock-solid, conflict-free offline delta synchronization mechanism that guarantees zero data loss or duplicate regulatory filings during intermittent connectivity [INFERENCE].

First 5 Actions

1. Define Core JSON Schemas: Map exact fields for DGMS Form 14, Form 15, and Mines Rules Form XIII [INFERENCE].
2. Initialize Flutter App Base: Implement local SQLite storage and WebP camera photo compression pipeline [INFERENCE].
3. Build Vector Clock Sync API: Create FastAPI sync endpoints with PostgreSQL schema support [INFERENCE].
4. Train Baseline Risk Model: Train XGBoost on synthetic inspection CSV logs to output pit risk scores [INFERENCE].
5. Develop React Dashboard Wireframe: Construct multi-tenant views for Colliery Managers and HQ Officers [INFERENCE].

Validation Before Coding

• Verify that all extracted form fields align strictly with Coal Mines Regulations 2017 requirements [RESEARCHED].
• Confirm mobile camera EXIF metadata extraction accurately captures GPS coordinates and hardware timestamps [INFERENCE].

Buildability Score /100

92/100 (Highly feasible for a skilled 4-6 student team using standard open-source web, mobile, and ML frameworks within the SIH timeframe) [INFERENCE].

Confidence

High [INFERENCE]. The proposed architecture strictly preserves all explicit requirements of SIH26024 while grounding technical choices in verifiable field realities of Coal India Limited [PS, RESEARCHED].

Research Sources

1. Ministry of Coal, Government of India
   • URL: https://coal.nic.in/ [1.9]
   • Type: Official Government Portal
   • Supported Claim: Administrative control, safety policies, and overarching governance of Coal India Limited [1.9].
   • Reliability: Official Primary Source
2. Coal India Limited (CIL) Official Website
   • URL: https://www.coalindia.in/ [3.10]
   • Type: Official Corporate Portal
   • Supported Claim: Operational structure, subsidiary management (BCCL, CCL, ECL, MCL, NCL, SECL, WCL, CMPDI) [1.2, 3.10].
   • Reliability: Official Primary Source
3. CIL Real Time Contravention Reporting System (ODRS)
   • URL: https://www.coalindiaodrs.live/ [1.6]
   • Type: Official System Portal
   • Supported Claim: Existing unsafe acts, conditions, and near-miss reporting mechanisms in CIL [1.6].
   • Reliability: Official Primary Source
4. Central Mine Planning & Design Institute (CMPDI) - Geomatics & Systems
   • URL: https://www.cmpdi.co.in/ [3.1]
   • Type: Official Subsidiary Portal
   • Supported Claim: CMSMS satellite monitoring system, Khanan Prahari mobile app, and GIS spatial capabilities [3.1, 3.2].
   • Reliability: Official Primary Source
5. Coal Mines Regulations 2017 Summary & Framework
   • URL: https://minemountain.in/mining/coal-mines-regulation-2017 [2.1]
   • Type: Educational / Legal Mining Resource
   • Supported Claim: CMR 2017 statutory safety requirements, Regulation 104 Safety Management Plans, and DGMS compliance frameworks [2.1].
   • Reliability: Reputable Domain Portal
6. DGMS Compliance Software & Form 14/15 Standards in Indian Mining
   • URL: https://iceipts.com/products/dgms-compliance [1.3]
   • Type: Industry Technical Portal
   • Supported Claim: Automation of Form 14, Form 15, Form XIII, and digital audit trail requirements under DGMS norms [1.3, 1.11].
   • Reliability: Secondary Technical Source