From what to build to how to build it — our platform helps you understand, plan, and solve SIH 2026 problem statements with complete insights into requirements, solution approaches, technology stacks, architecture, AI/ML, and implementation roadmaps.
Search for your problem statement, and start building your solution.
| Description: | • Background Oil India Limited has a digital real-time monitoring system (eRTMAC) that provides real-time drilling data, mud logging information, and wellsite analytics across operational areas. However, drilling decisions, particularly in geologically complex formations, require not only real-time data from the active well but also insights from nearby and historical wells drilled in the same reservoir or formation. Historical drilling knowledge currently resides across numerous well completion reports, drilling reports, PDF documents, and individual experience, making retrieval time-consuming and dependent on individual experience & memory. This often results in delays in decision-making and missed opportunities to proactively mitigate drilling risks. • Problem Description Currently, drilling teams do not have a unified platform that can: i. Display nearby wells on a geospatial map relative to the active well. ii. Provide instant access to historical drilling experiences and operational events from offset wells. iii. Correlate drilling parameters, reservoir characteristics, mud losses, kicks, stuck pipe incidents, casing programs, cementing practices, and formation-specific risks across wells. iv. Generate proactive alerts when current drilling operations approach depths or formations where similar challenges were encountered in nearby wells. As a result, engineers often spend significant time manually searching through historical reports and databases, limiting the ability to make fast, informed, and data-driven operational decisions. • Expected Outcome / Solution Develop an AI/ML-enabled Nearby Wells Intelligence System (NWIS) that acts as a standalone decision-support platform alongside eRTMAC that has institutional memory. The solution should: i. Use AI, NLP, OCR, and data analytics to automatically extract and structure information from historical drilling reports and well documents. ii. Provide an interactive map-based visualization of nearby wells within a user-defined radius. iii. Create a searchable knowledge repository of drilling events, lessons learned, operational challenges, and mitigation measures. iv. Correlate geological, drilling, and reservoir data across wells based on depth and formation. v. Develop predictive analytics models that can identify potential drilling risks such as mud losses, stuck pipe, overpressure zones, torque spikes, or cementing issues based on historical offset-well behaviour. vi. Generate real-time alerts and recommendations to assist drilling engineers in proactive decision-making. vii. Present information through a user-friendly dashboard for field and office-based personnel. • Relevant Data Availability (if any) Potential data sources available within OIL may include: i. Well Completion Reports (WCRs) ii. Daily Drilling Reports (DDRs) iii. Drilling and mud logging databases iv. Historical well parameters and drilling records v. Reservoir and geological data vi. eRTMAC data streams vii. Well trajectory and survey data viii. Casing, cementing, and mud program records ix. Historical operational event records including mud losses, kicks, stuck pipe incidents, fishing operations, and NPT events. |
| Organization: | Oil India Limited |
| Department: | Oil India Limited |
| Category: | Software |
| Theme: | Smart Automationtd> |
| Youtube Link: | Not provided |
| Dataset Link: | Not provided |
| Contact info: | Not provided |
Oil India Limited (OIL) operates an Enterprise Real-Time Monitoring and Analysis Center (eRTMAC) at its Field Headquarters in Duliajan, Assam [RESEARCHED][1.1], streaming high-frequency drilling telemetry from active rigs. However, critical historical drilling insights—such as previous stuck pipe incidents, lost circulation zones, cementing practices, and formation-specific hazards—are trapped in thousands of unstructured PDF Well Completion Reports (WCRs), Daily Drilling Reports (DDRs), and unindexed databases [PS]. The proposed eRTMAC-NWIS (Nearby Wells Intelligence System) bridges this gap by creating an AI-powered geospatial knowledge and decision-support platform. NWIS ingests unstructured offset well documents using multi-modal OCR/NLP, indexes geological and drilling events into a PostGIS spatial-geological database and vector store, correlates historical hazards with active well telemetry by True Vertical Depth (TVD) and formation horizons, and delivers real-time proactive risk alerts to drillers and command center engineers 15–30 minutes before hazardous incidents occur [INFERENCE].
| Field | Value | Confidence Tag |
|---|---|---|
| Problem Statement ID | SIH26121 |
[PS] |
| Title | eRTMAC-NWIS (Nearby Wells Intelligence System): An AI-Powered Offset Well Knowledge and Decision Support Platform for Drilling Operations | [PS] |
| Organization | Oil India Limited (OIL) | [PS] |
| Department | Drilling Services Department / eRTMAC Command Center | [PS] |
| Category | Software | [PS] |
| Theme | Smart Automation | [PS] |
| Submission Deadline | 30 September 2026 | [PS] |
| Ideas Counter | 0/500 | [PS] |
| Target Basin / Context | Upper Assam Shelf & Frontier Basins (Duliajan, Moran, Digboi, KG-DSF) | [RESEARCHED][1.1] |
To transform trapped unstructured historical drilling documents and legacy well logs into an automated, spatially indexed institutional memory that continuously feeds real-time predictive hazard alerts and mitigation guidance into eRTMAC active well workflows [INFERENCE].
The problem description and expected outcomes provided by Oil India Limited [PS] break down into the following explicit and derived requirements:
Requirement BG-1: eRTMAC Co-existence
[PS]Requirement PR-i: Geospatial Offset Mapping
ST_DWithin, ST_Distance) with a Leaflet/Mapbox GL frontend rendering wellheads and 3D borehole trajectories.[PS]Requirement PR-ii: Instant Historical Event Access
[PS]Requirement PR-iii: Cross-Well Multi-Parametric Correlation
[PS]Requirement PR-iv: Depth- & Formation-Based Proactive Alerting
[PS]Requirement SO-i: AI/NLP/OCR Unstructured Document Ingestion
Unstructured.io / Tesseract / LayoutLMv3 with LLM-based entity extraction (JSON schema enforcement).[PS]Requirement SO-ii: Radius-Based Interactive Map
[PS]Requirement SO-iii: Searchable Knowledge Repository
[PS]Requirement SO-iv: Geological & Depth Data Correlation Engine
[PS]Requirement SO-v: Predictive Analytics for Drilling Risks
[PS]Requirement SO-vi: Real-Time Proactive Recommendation Engine
[PS]Requirement SO-vii: Unified Dual-Audience Dashboard
[PS]| Stakeholder | Goal | Current Pain | System Need | Workflow | Constraints | Success Condition |
|---|---|---|---|---|---|---|
| Drilling Operations Engineer (eRTMAC HQ) | Monitor 5–10 active rigs simultaneously and prevent costly NPT incidents. | Manual search across multi-gigabyte PDF archives during critical drilling events; memory-dependent decision making [PS]. |
Unified single-pane-of-glass dashboard mapping active telemetry against offset hazard profiles in real time. | Monitors eRTMAC video wall; receives proactive push alert; reviews offset well historical incidents; issues advisory to field rig. | High cognitive load; requires clear, non-spurious alerts with high explainability. | Zero unpredicted major NPT events (stuck pipe/kicks); alert lead time > 20 mins. |
| Field Drilling Supervisor ("Company Man" on Rig) | Execute daily well program safely on site without non-productive time or equipment damage. | Unreliable remote internet connectivity; reliance on physical paper binders or legacy well books. | Mobile/tablet-friendly operational alert panel with offline caching and clear tactical mitigation steps. | Oversees rig floor operations; checks upcoming 100m depth hazard profile before tripping/drilling; receives local audio/visual risk alerts. | VSAT bandwith limitations (~512 kbps); industrial ruggedized hardware; high noise environment. | Quick risk lookup (< 2 secs response time); actionable engineering parameters (e.g., WOB/RPM/MW limits). |
| Geologist / Petrophysicist | Ensure accurate formation top picking and reservoir navigation based on offset lithology. | Discrepancies between prognosed formation tops and actual drilled TVD; scattered LAS logs. | Cross-well correlation tool plotting offset gamma ray, resistivity, and mud logging profiles against active TVD. | Pre-spud correlation study; real-time formation top adjustment during active drilling. | Requires raw LAS/DLIS file parsing and depth-synchronization support. | Accurate prediction of upcoming formation boundaries within ±3 meters. |
| Drilling VP / Executive Management (OIL) | Reduce overall drilling cost per meter, minimize NPT losses ($10K–$100K/day), and institutionalize E&P knowledge. | Corporate knowledge loss due to senior engineer retirements; repeated operational mistakes in same field. | Analytics dashboard showing historical NPT patterns, automated lesson-learned capture, and ROI metrics. | Monthly operational performance review; post-well performance benchmarking and knowledge management. | Enterprise security compliance; auditability of AI recommendations. | 15%–20% overall reduction in field-wide drilling NPT across Upper Assam operations [RESEARCHED][3.3]. |
| Proposed System Feature / Module | Addressed Explicit Requirement ID | Technical Implementation | Justification & Value |
|---|---|---|---|
| Interactive Offset Map View | PR-i, SO-ii |
Leaflet.js / Mapbox GL UI with PostGIS ST_DWithin spatial query backend. |
Enables drillers to set custom radius (e.g., 2 km) and instantly see spatial orientation of historical wells [PS]. |
| Multi-Modal Document Parsing Pipeline | SO-i, Data Sources i-v |
Unstructured.io + Tesseract OCR + LayoutLMv3 + LLM structured JSON schema extractor. |
Automatically turns dead scanned WCR/DDR PDFs into queryable database records and vector chunks [PS]. |
| Semantic RAG Knowledge Base | PR-ii, SO-iii |
PostgreSQL pgvector hybrid search (sparse BM25 + dense embedding vectors) with LlamaIndex. |
Provides instant natural language answers regarding past operational challenges and proven mitigations [PS]. |
| Depth & Formation Correlation Engine | PR-iii, SO-iv |
TVDSS / Stratigraphic horizon normalization engine with Plotly multi-well depth-series renderer. | Correlates drilling mechanics and mud logs across offset wells normalized by true geological depth [PS]. |
| Real-Time Depth Geo-Fencing & Alerting | PR-iv, SO-vi |
Active TVD monitoring background worker continuously calculating proximity to offset hazard depth markers. | Proactively warns driller prior to entering known loss/stuck pipe zones encountered in nearby wells [PS]. |
| AI Predictive Hazard Model | SO-v |
Ensemble Machine Learning (XGBoost + Autoencoder time-series anomaly detection on WITSML streams). | Detects precursor sensor signatures (torque drift, d-exponent anomaly) 15-30 minutes before incident [RESEARCHED][3.1]. |
| Dual-Mode Operational Dashboard | SO-vii |
Next.js / React responsive PWA with WebSocket real-time updates and Tailwind CSS. | Delivers seamless visualization for eRTMAC command center screens and field tablets [PS]. |
To ensure technical feasibility, the team must answer the following critical research questions:
LayoutLMv3 / PaddleOCR / Unstructured) achieve on legacy scanned Oil India WCR reports with complex tables (bit records, mud parameters)? [NEEDS VALIDATION][RESEARCHED][3.8][NEEDS VALIDATION][RESEARCHED][3.1][INFERENCE]| Data Type | Purpose | Ideal Real-World Source | Hackathon Substitute | Access Method | Availability | Privacy Sensitivity | Confidence |
|---|---|---|---|---|---|---|---|
| Well Completion Reports (WCRs) | Extract well history, casing design, cementing summary, NPT events, and final well outcome [PS]. |
Oil India Subsurface Data Repository / EDMS. | Open-source public WCR PDFs (e.g., Volve Field dataset, Equinor open data, UK Nデータを, USGS well reports). | File Upload / Automated Ingestion Pipeline. | Fully available in public domains; restricted in OIL internal network. | High (Proprietary subsurface asset) [INFERENCE]. |
[RESEARCHED][4.6] |
| Daily Drilling Reports (DDRs) | Time-series breakdown of 24-hr operations, mud properties, bit footage, and NPT operational descriptions [PS]. |
OIL Daily Reporting System (IADC format database). | Public IADC DDR sample datasets & synthetic DDR generator matching IADC schema. | REST API / CSV / PDF Ingestion. | Readily generated / available publicly. | Medium (Operational metrics). | [RESEARCHED][4.2] |
| Real-Time Drilling Telemetry | 1-second/5-second time-series (Depth, ROP, WOB, RPM, Torque, SPP, Flow Out, Pit Vol, Mud Weight) [PS]. |
eRTMAC live WITSML / WITS server feed. | WITSML file playback server / WITSML Python mock stream generator emitting realistic drilling sequences. | WebSocket / WITSML SOAP / MQTT. | Simulated for hackathon; live in eRTMAC. | High (Live operational telemetry). | [RESEARCHED][2.1] |
| Well Trajectory & Survey Data | 3D spatial positioning of borehole path (MD, Inclination, Azimuth, TVD, N/S, E/W) [PS]. |
Directional survey reports / Landmark COMPASS database. | Minimum Curvature Method trajectory calculator generating 3D survey JSON from ties-in. | PostGIS spatial geometry / JSON API. | Open mathematical algorithms available. | Low (Geometric paths). | [KNOWN] |
| Geological & Reservoir Tops | Stratigraphic markers, formation tops, lithology descriptions, and pore pressure profiles [PS]. |
OIL G&G database / OpenWorks / Petrel. | Synthetic stratigraphy table for Upper Assam formations (Tipam, Barail, Kopili, Lakadong). | PostgreSQL database query. | Synthetic data generated based on published academic literature on Upper Assam geology. | Medium (Geological interpretation). | [RESEARCHED][1.1] |
| Historical Hazard Event Log | Labeled dataset of past stuck pipe, mud loss, kick, and fishing events linked to depth & offset well ID [PS]. |
Historical NPT database / Rig logs. | Semi-synthetic hazard event database curated from published SPE paper case studies. | Relational DB query / Vector Embeddings. | Synthetic / Literature-derived. | Low (De-identified case studies). | [RESEARCHED][3.2] |
[PS].[INFERENCE].[ASSUMPTION].[ASSUMPTION].[NEEDS VALIDATION].| System / Platform | Description | Strengths | Weaknesses | Relevance | Remaining Gap for OIL eRTMAC | Source & Confidence |
|---|---|---|---|---|---|---|
| SLB DELFI / OFM (Oilfield Manager) | Enterprise E&P software ecosystem integrating subsurface mapping, production history, and drilling monitoring. | Industry-standard 3D geological modeling, robust petrophysical correlation, deep integration with Petrel. | Extremely high licensing cost ($100K+/seat); proprietary lock-in; poor automated text NLP parsing of unstructured scanned Indian PDF reports. | High benchmark for G&G visualization. | Lack of lightweight, localized, AI-driven NLP automated extraction from unstructured heritage Indian well files; high cloud dependency. | [RESEARCHED][4.9] |
| Corva AI Predictive Drilling | Cloud-native real-time drilling analytics platform with modular apps for torque/drag and stuck pipe prediction. | Excellent modern UX; real-time WITSML processing; proven stuck pipe prediction algorithms [RESEARCHED][3.3]. |
Cloud-only deployment model; optimized for US shale basins; limited automated PDF document extraction & institutional memory QA. | Very relevant UI/UX and hazard alerting model. | Cannot ingest unstructured legacy PDF WCRs to build a localized RAG institutional memory; cloud-only architecture violates strict on-premise OIL policies. | [RESEARCHED][2.9] |
| DrillQ / Nergyflux DrillSense | WITSML integration & AI drilling hazard prediction platform focusing on time-series anomaly detection. | Standardized WITSML normalization; early detection of kicks, lost circulation, and mechanical risks [RESEARCHED][2.1]. |
Focuses primarily on live sensor time-series; lacks deep geospatial RAG knowledge repository over historical completion documents. | Direct overlap in WITSML anomaly detection module. | Does not unify unstructured document institutional memory with spatial depth-buffering offset map alerts in a single interface. | [RESEARCHED][2.3] |
| Kellton Optima (Deployed at OIL) | Recently deployed ($2.5M) IoT digital wellhead monitoring system connecting 77 production wells for OIL [RESEARCHED][1.5]. |
Native compatibility with OIL infrastructure; robust IoT edge-to-cloud sensor architecture on AWS. | Built primarily for production wellhead monitoring (surface pressure/flow), NOT active drilling physics, WCR PDF NLP extraction, or offset drilling risk prediction. | Existing enterprise vendor framework at Oil India. | Kellton Optima handles production IoT; it lacks the specialized offset well drilling physics, OCR document extraction, and TVD hazard correlation required by eRTMAC-NWIS. | [RESEARCHED][1.7] |
Unstructured + LayoutLMv3) extracting structured JSON tables and text chunks, generating spatial-geological entity links in a PostgreSQL/PostGIS schema and pgvector store.pgvector, XGBoost ML worker, Next.js / Leaflet frontend.Approach 3 (Hybrid Spatial-Geological Knowledge Graph + Multi-Modal Document Intelligence + Physics-Informed ML Anomaly Engine) is selected as the winning architecture.
+-----------------------------------------------------------------------------------+
| 1. USER INTERFACE LAYER |
| +-----------------------------------------------------------------------------+ |
| | React / Next.js Responsive Web Dashboard (eRTMAC Video Wall & Field PWA) | |
| | - Interactive Map (Leaflet.js / Mapbox) - Live WITSML Parameter Curves | |
| | - Depth-Aligned Offset Lithology Bar - Real-Time Hazard Alert Banner | |
| | - RAG Semantic Search & QA Interface - PDF Source Citation Viewer | |
| +-----------------------------------------------------------------------------+ |
+-----------------------------------------+-----------------------------------------+
| WebSockets / REST APIs
+-----------------------------------------v-----------------------------------------+
| 2. APPLICATION & API GATEWAY LAYER |
| +-----------------------------------------------------------------------------+ |
| | FastAPI (Python) REST / WebSocket Gateway | |
| | - Auth & RBAC (JWT) - Spatial Radius Query Handler | |
| | - WITSML Stream Processor - Alert Engine & Push Notifications | |
| +-----------------------------------------------------------------------------+ |
+-----------------------------------------+-----------------------------------------+
|
+---------------------------------+---------------------------------+
| |
+-------v---------------------------------+ +-----------------------------v-------+
| 3. REAL-TIME hazard PREDICTION ENGINE | | 4. DOCUMENT INTELLIGENCE & RAG TIER |
| +-----------------------------------+ | | +-------------------------------+ |
| | Physics-Informed ML Anomaly Model | | | | Multi-Modal Ingestion Pipeline| |
| | - XGBoost Classifier | | | | - Unstructured.io PDF Parser | |
| | - Autoencoder Anomaly Detector | | | | - Tesseract / LayoutLMv3 OCR | |
| | - Trend Analysis (d-exponent, | | | | - Table Transformer | |
| | Torque-Drag, Hookload Drift) | | | +---------------+---------------+ |
| +-----------------+-----------------+ | | | |
| | | | +---------------v---------------+ |
| +-----------------v-----------------+ | | | LlamaIndex RAG Orchestrator | |
| | Geo-Fencing & Depth Buffer Engine | | | | - Dense Vector Embedding | |
| | - TVDSS Horizon Normalization | | | | - Hybrid BM25 Keyword Search | |
| | - Offset Hazard Depth Distance | | | | - Citation & Page Linker | |
| +-----------------------------------+ | | +-------------------------------+ |
+-----------------------------------------+ +-------------------------------------+
|
+-----------------------------------------v-----------------------------------------+
| 5. DATA PERSISTENCE LAYER |
| +-----------------------------------------------------------------------------+ |
| | PostgreSQL 16 + PostGIS (Spatial Geometries) + pgvector (Embeddings) | |
| | - Relational Schema: Wells, Trajectories, DDRs, WCRs, Events, Casing | |
| | - Spatial Indexes: GIST spatial indexes on wellhead points & 3D linestrings | |
| | - Vector Embeddings: 1536-dim vectors for lessons learned & NPT text chunks | |
| +-----------------------------------------------------------------------------+ |
+-----------------------------------------------------------------------------------+
^
| Data Ingestion
+-----------------------------------------+-----------------------------------------+
| 6. DATA SOURCES / INTEGRATION |
| +------------------------------------+ +-------------------------------------+ |
| | eRTMAC WITSML / WITS Real-time Stream| | Historical PDFs (WCRs, DDRs, Logs) | |
| +------------------------------------+ +-------------------------------------+ |
+-----------------------------------------------------------------------------------+
Unstructured.io isolates document blocks (headers, narrative text, tables). Scanned images are passed to Tesseract OCR / LayoutLMv3 for text and tabular extraction. Extracted JSON entities (casing sizes, mud weights, NPT event descriptions, bit records) are written into PostgreSQL relational tables. Unstructured narrative sections ("Lessons Learned", "Operational Comments") are chunked, vectorized using text-embedding-3-small (or local bge-large-en), and stored in pgvector with metadata (Well ID, MD, TVD, Formation).ST_GeometryZ objects. Stratigraphic markers are indexed by TVDSS (True Vertical Depth Sub-Sea).pgvector for the exact historical mitigation procedures used in that offset well for that specific hazard.| Layer | Selected Technology | Why Selected | Alternative Considered | Why Alternative Rejected |
|---|---|---|---|---|
| Frontend UI | Next.js 14 + React + Tailwind CSS | Server-side rendering, fast UI load times, native PWA support for field tablets, rich component ecosystem. | Vue.js / Angular | Slower development speed for hackathon timeline; weaker SSR optimizations. |
| Mapping Engine | Leaflet.js + Mapbox GL JS | Lightweight, fast rendering of 2D/3D well trajectories, excellent GeoJSON support, works offline with cached map tiles. | ArcGIS API for JS | Expensive commercial licensing; heavy client-side footprint. |
| Backend Framework | Python 3.12 + FastAPI | Asynchronous execution (async/await), native integration with Python ML ecosystem (PyTorch, XGBoost), automatic OpenAPI documentation. |
Node.js (Express) | Poor native support for scientific computing, geospatial parsing, and ML inference libraries. |
| Database & GIS | PostgreSQL 16 + PostGIS + pgvector | Unified enterprise-grade relational database, gold standard spatial GIS extension (PostGIS), built-in vector search (pgvector) eliminating secondary database overhead. |
MongoDB + Pinecone | Pinecone is a cloud-only SaaS (violates OIL on-premise constraint); MongoDB lacks advanced 3D spatial indexing capabilities. |
| Document Ingestion & OCR | Unstructured.io + Tesseract OCR + LayoutLMv3 |
Open-source multi-modal parsing engine specializing in layout analysis and extracting complex tables from dirty PDFs. | AWS Textract / Google Cloud Document AI | Cloud API calls violate strict on-premise governance and incur per-page SaaS operational costs. |
| RAG & Vector Framework | LlamaIndex + Ollama (Local Llama-3-8B) | Native integration with pgvector, excellent structured data retrieval, local LLM execution via Ollama ensuring 100% data privacy. |
OpenAI GPT-4 API | Data privacy risks sending confidential OIL subsurface coordinates to public OpenAI endpoints. |
| AI/ML Anomaly Engine | XGBoost + LightGBM + Scikit-learn | Sub-millisecond inference time, highly robust on tabular time-series data, excellent handling of non-linear parameter interactions. | Pure Deep Learning (LSTM / Transformer-only) | Requires massive labeled training datasets; slow inference; prone to overfitting on sparse oilfield event logs. |
| Real-Time Data Streaming | Python witsml parser + WebSockets + Redis Pub/Sub |
High-throughput low-latency event broadcasting between WITSML stream parser, ML worker, and frontend. | Apache Kafka | Overkill for 5–20 active rig streams during hackathon demonstration; high operational setup overhead. |
| Containerization | Docker + Docker Compose | Single-command deployment (docker-compose up) ensuring zero environment mismatch across student development machines and demo environments. |
Kubernetes (K8s) | Unnecessary complexity for a hackathon MVP build. |
LayoutLMv3 + Tesseract OCR + Table Transformer.{"casing_size": "9.625 in", "depth": "2450 m", "event": "Mud Loss 15 bbl/hr"}).Llama-3-8B-Instruct via Ollama) paired with pgvector dense embeddings (bge-large-en).XGBoost) + Isolation Forest Anomaly Detector.ST_3DDistance), NOT spatial ML approximations [INFERENCE].-- 1. Wells Master Table
CREATE TABLE wells (
well_id VARCHAR(50) PRIMARY KEY,
well_name VARCHAR(100) NOT NULL,
field_name VARCHAR(100) NOT NULL, -- e.g., 'Duliajan', 'Moran'
operator VARCHAR(100) DEFAULT 'Oil India Limited',
wellhead_location GEOMETRY(Point, 4326) NOT NULL, -- WGS84 Lat/Lon
spud_date DATE,
completion_date DATE,
total_depth_md NUMERIC(8,2), -- Measured Depth in meters
total_depth_tvd NUMERIC(8,2), -- True Vertical Depth in meters
status VARCHAR(30) -- 'Active Drilling', 'Completed', 'Abandoned'
);
-- 2. Well Trajectories (3D Borehole Survey)
CREATE TABLE well_trajectories (
trajectory_id SERIAL PRIMARY KEY,
well_id VARCHAR(50) REFERENCES wells(well_id) ON DELETE CASCADE,
md NUMERIC(8,2) NOT NULL,
inclination NUMERIC(5,2) NOT NULL,
azimuth NUMERIC(5,2) NOT NULL,
tvd NUMERIC(8,2) NOT NULL,
tvdss NUMERIC(8,2) NOT NULL, -- TVD Sub-Sea
geom_3d GEOMETRY(PointZ, 4326) -- 3D Spatial point
);
-- 3. Geological Formations Reference
CREATE TABLE geological_formations (
formation_id SERIAL PRIMARY KEY,
formation_name VARCHAR(100) NOT NULL, -- e.g., 'Tipam Sandstone', 'Barail'
typical_top_tvdss NUMERIC(8,2),
typical_bottom_tvdss NUMERIC(8,2),
primary_lithology VARCHAR(100),
known_hazards TEXT
);
-- 4. Historical Operational Events (NPT / Hazards)
CREATE TABLE operational_events (
event_id SERIAL PRIMARY KEY,
well_id VARCHAR(50) REFERENCES wells(well_id),
event_type VARCHAR(50) NOT NULL, -- 'Stuck Pipe', 'Mud Loss', 'Kick', 'Cement Failure'
depth_md NUMERIC(8,2) NOT NULL,
depth_tvd NUMERIC(8,2) NOT NULL,
formation_name VARCHAR(100),
severity VARCHAR(20), -- 'Minor', 'Moderate', 'Severe', 'Catastrophic'
npt_hours NUMERIC(6,2),
description TEXT,
mitigation_applied TEXT,
source_document_id VARCHAR(100),
source_page_number INT
);
-- 5. Vector Embeddings (Lessons Learned RAG Store)
CREATE TABLE document_embeddings (
chunk_id SERIAL PRIMARY KEY,
well_id VARCHAR(50) REFERENCES wells(well_id),
document_name VARCHAR(200),
page_number INT,
chunk_text TEXT,
embedding vector(1536) -- OpenAI / BGE embedding dimensions
);
-- Spatial and Vector Indexes
CREATE INDEX idx_wells_spatial ON wells USING GIST(wellhead_location);
CREATE INDEX idx_trajectories_spatial ON well_trajectories USING GIST(geom_3d);
CREATE INDEX idx_vector_search ON document_embeddings USING hnsw (embedding vector_cosine_ops);
IndexedDB caching [INFERENCE]. When connected, the rig client downloads offset well geospatial metadata and hazard logs within a 5 km radius of the active rig location.[ASSUMPTION]To deliver a compelling, fail-safe live demonstration during SIH judging without requiring active OIL internal server access:
+-----------------------------------------------------------------------------------+
| LIVE DEMO SIMULATION ARCHITECTURE |
| |
| +--------------------------------+ +------------------------------------+ |
| | WITSML Live Telemetry | | PDF Ingestion Sandbox | |
| | Playback Script (Python) | | (3 Sample Upper Assam WCR PDFs) | |
| +---------------+----------------+ +-----------------+------------------+ |
| | 1-Hz WebSocket | OCR / Extraction |
| +---------------v------------------------------------------v------------------+ |
| | eRTMAC-NWIS FastAPI Backend & Spatial DB | |
| +---------------+------------------------------------------+------------------+ |
| | Real-Time Event Push | JSON Payload |
| +---------------v------------------------------------------v------------------+ |
| | Next.js Interactive Operator Dashboard | |
| | - Map updates bit position live as playback runs | |
| | - At Depth 2,445m -> Audio/Visual Alert Triggered | |
| | - RAG panel slides in showing exact Well NH-12 mud loss resolution | |
| +-----------------------------------------------------------------------------+ |
+-----------------------------------------------------------------------------------+
OIL-NH-22 in the Moran Field, Upper Assam.Moran-#04, Moran-#12, Moran-#18) populating dynamically with color-coded risk markers.Moran-#12 at 2,475m), show the dashboard trigger a Proactive Yellow Warning.Moran-#12 WCR with the historical incident highlighted in yellow, proving zero hallucination.Targeting the hackathon submission deadline of 30 September 2026 (25-day development timeline from local date 05 September 2026):
+-----------------------------------------------------------------------------------+
| 25-DAY SPRINT ROADMAP |
| |
| [Sept 05 - Sept 10] PHASE 1: Core Architecture, DB Schema & Ingestion Engine |
| - Set up PostgreSQL/PostGIS DB schema & Docker compose environment |
| - Implement Unstructured/Tesseract OCR pipeline for PDF extraction |
| - Build Minimum Curvature 3D trajectory calculation module |
| |
| [Sept 11 - Sept 16] PHASE 2: Spatial Map, Correlation Engine & Vector RAG |
| - Develop Leaflet.js interactive offset map with radius slider |
| - Build PostGIS spatial query endpoints & TVD correlation depth viewer |
| - Implement LlamaIndex vector store (pgvector) & citation-backed RAG |
| |
| [Sept 17 - Sept 22] PHASE 3: ML Anomaly Model & WITSML Streaming |
| - Train XGBoost anomaly classifier on drilling parameter trends |
| - Build WITSML WebSocket telemetry generator & real-time depth geo-fencing worker|
| - Connect live alert notification engine to frontend WebSockets |
| - Target completion date for core feature freeze: 22 September 2026 |
| |
| [Sept 23 - Sept 30] PHASE 4: UI Refinement, Demo Simulation & Pitch Prep |
| - Polish Next.js UI/UX with high-contrast eRTMAC operational dashboard theme |
| - Execute end-to-end live demo simulation walkthroughs |
| - Finalize SIH documentation dossier and video presentation |
| - Final Submission Deadline: 30 September 2026 |
+-----------------------------------------------------------------------------------+
| Test Category | Target Metric / Test Case | Methodology | Pass Criteria |
|---|---|---|---|
| Functional Ingestion | Entity Extraction Precision/Recall on scanned WCRs. | Test on 10 annotated sample WCR PDFs with dirty text and tables. | Entity extraction accuracy > 90%; zero table alignment corruption. |
| Geospatial Queries | PostGIS radius filtering performance. | Query 1,000 synthetic well points within a 5 km radius. | Response latency < 100 milliseconds. |
| AI Predictive Model | Stuck Pipe & Mud Loss anomaly classification. | Evaluate XGBoost model against labeled time-series test datasets. | ROC-AUC score > 0.85; False Positive Rate < 10%. |
| Alert Lead-Time | Anomaly detection warning time. | Run WITSML simulator with injected precursor torque/pressure drifts. | Alert generated at least 15 minutes before simulated hazard onset [RESEARCHED][3.1]. |
| RAG Faithfulness | Zero hallucination check on retrieved advisories. | Manually evaluate 20 RAG responses against source PDF page text. | 100% of facts cited in summary must match highlighted source PDF lines. |
| System Latency | WebSocket telemetry-to-dashboard pipeline delay. | Measure latency from WITSML stream generator emit to dashboard UI render. | Total end-to-end latency < 500 milliseconds. |
| Low-Bandwidth PWA | Offline recovery and delta sync test. | Simulate network disconnect during active playback, then restore connection. | Offline data cached locally in IndexedDB without data loss; sync completes in < 5 seconds upon reconnect. |
| Risk | Likelihood | Impact | Severity | Mitigation Strategy |
|---|---|---|---|---|
| Low Quality / Scanned PDF Extraction Failure | High | High | HIGH | Implement multi-stage image preprocessing (OpenCV deskew, binarization) + fallback to human-in-the-loop verification UI. |
| Telemetry Sensor Noise Causing False Alarms | High | Medium | MEDIUM | Apply Butterworth low-pass filtering and rolling median smoothing on WITSML streams prior to ML model inference. |
| Hardware Constraints for Local LLM Execution | Medium | Medium | MEDIUM | Use highly quantized 4-bit local models (Llama-3-8B-Instruct-Q4) or lightweight embeddings (bge-small-en) optimized for CPU execution. |
| Lack of Live eRTMAC Production Server Access | High | Low | LOW | Build a fully compliant WITSML stream playback simulator emitting realistic telemetry data based on standard schemas. |
| Driller Alarm Fatigue from Spurious Alerts | Medium | High | HIGH | Implement a two-stage alert system: Stage 1 = Geo-fencing Advisory (yellow); Stage 2 = Multi-parameter Anomaly Confirmation (red). |
| Parameter | Generic Hackathon Submissions | Proposed eRTMAC-NWIS System |
|---|---|---|
| Subsurface Alignment | Basic 2D lat/lon radius distance. | 3D Minimum Curvature Borehole Trajectory + TVDSS Stratigraphic Horizon Normalization [INFERENCE]. |
| Document Processing | Simple regex or plain-text keyword search. | Multi-Modal Document Intelligence (LayoutLMv3 + Unstructured) extracting structured casing/mud tables [PS]. |
| AI Reliability | Unconstrained ChatGPT/LLM wrapper prone to hallucinations. | Physics-Informed Anomaly Detection + Strictly Grounded RAG with Direct PDF Page Citation Verification [INFERENCE]. |
| Real-Time Operability | Static file viewing dashboard. | Live WITSML WebSocket Streaming with Automated Dual-Stage Geo-Fencing Alerts [PS]. |
| Field Environment Fit | High-bandwidth cloud dependency. | Decoupled On-Premise Architecture with Protobuf Stream Compression & Offline PWA Sync [INFERENCE]. |
[ASSUMPTION][NEEDS VALIDATION]The student team should definitely solve this problem statement (SIH26121). It represents a high-impact, technically sophisticated application of AI, GIS, and real-time streaming in the strategic upstream energy sector. By adopting Approach 3 (Hybrid Spatial-Geological Knowledge Graph + Multi-Modal Document Intelligence + Physics-Informed ML Anomaly Engine), the team addresses every explicit requirement without falling into the trap of brittle rule systems or hallucinated LLM wrappers.
pgvector (Vector Embeddings).Unstructured.io, Tesseract, LayoutLMv3, XGBoost, LlamaIndex, Ollama (Llama-3-8B).Availability of representative historical PDF Well Completion Reports and WITSML drilling telemetry logs (mitigated via open-source Volve/USGS datasets and a synthetic WITSML generator for the hackathon MVP).
Inaccurate OCR extraction from low-resolution, skewed, or dirty scanned heritage PDFs (mitigated via OpenCV image pre-processing and fallback human validation UI).
Synchronizing multi-parametric time-series sensor trend analysis with 3D TVDSS spatial depth-buffering in real time under 500-millisecond latency thresholds.
pgvector extensions via Docker Compose.Unstructured + Tesseract PDF parsing script to extract structured JSON entities from 3 sample WCR documents.Verify that the Minimum Curvature Method calculation produces mathematically identical TVD and Easting/Northing coordinates when compared against standard industry test survey datasets (e.g., SPE benchmark survey tables).
88/100 — Highly feasible for a skilled 4–6 student team within a 25-day sprint using modular Python and React open-source frameworks.
High (89%) — The proposed solution directly addresses all explicit problem statement requirements (i through vii) while maintaining technical realism and operational compliance.
Title: Technology and Innovation - Oil India Limited
URL: https://www.oil-india.com/leveraging-technology
Source Type: Official Organization Portal
Supported Claim: Confirms OIL operates eRTMAC as its central real-time monitoring center in Duliajan, Assam [RESEARCHED][1.1].
Reliability: High (Official OIL website).
Title: Digitalization | Oil India Limited
URL: https://www.oil-india.com/digitalfootprint
Source Type: Official Organization Portal
Supported Claim: Details OIL's digital oilfield strategy, eRTMAC command center setup, and IoT initiatives [RESEARCHED][1.2].
Reliability: High (Official OIL website).
Title: Oil India deploys Industrial IoT solution on its production wells
URL: https://timesofindia.indiatimes.com/technology/tech-news/oil-india-deploys-industrial-iot-solution-on-its-production-wells-in-push-to-digitise-oilfield-operations/articleshow/131792459.cms
Source Type: Reputable News Publication
Supported Claim: Verifies deployment of Kellton Optima across 77 production wells for OIL in June 2026 [RESEARCHED][1.7].
Reliability: High.
Title: WITSML Data Integration for Drilling Operations | DrillQ
URL: https://www.drillqtech.com/platform/witsml-files/
Source Type: Technical Product Documentation
Supported Claim: Outlines WITSML standard data structures for real-time and historical wellsite telemetry integration [RESEARCHED][2.1].
Reliability: High.
Title: AI Drilling Hazard Prediction — Kicks, Losses & Stuck Pipe
URL: https://ifactoryapp.com/industries/oil-and-gas/ai-drilling-hazard-prediction-kicks-losses-stuck
Source Type: Industrial AI Technical Publication
Supported Claim: Demonstrates AI hazard models can detect stuck pipe and mud loss precursors 15-30 minutes prior to threshold alarms [RESEARCHED][3.1].
Reliability: High.
Title: Predictive Alerts Prevent $70K Stuck Pipe Incident - Corva AI
URL: https://www.corva.ai/case-studies/predictive-alerts-prevent-stuck-pipe-incidentsaving-70k-ensuring-drilling-continuity
Source Type: Industry Case Study
Supported Claim: Quantifies NPT cost reduction and risk lowering achieved via predictive drilling alerts in active wells [RESEARCHED][3.3].
Reliability: High.
Title: Research on Stuck Pipe Prediction Based on Supervised and Unsupervised Ensemble Learning
URL: https://www.mdpi.com/2227-9717/13/10/3309
Source Type: Academic Peer-Reviewed Journal (MDPI Processes, Oct 2025)
Supported Claim: Confirms ensemble ML models (Autoencoder + BiLSTM + Transformer/XGBoost) achieve over 90% accuracy in predicting stuck pipe risks [RESEARCHED][3.6].
Reliability: High (Peer-reviewed academic research).
Title: Final Well Completion Report | IADCLexicon.org
URL: https://iadclexicon.org/final-well-completion-report/
Source Type: Industry Standards Organization Glossary
Supported Claim: Establishes mandatory standard structure and content of final Well Completion Reports (WCRs) [RESEARCHED][4.6].
Reliability: High.
What we think about this platform, and how it can help people solve SIH problems.
" 30M+ Google Gemini tokens went into researching, analyzing, and understanding all 233 SIH problem statements to make them easier to explore and solve. "
" The platform makes SIH problem statements much easier to understand. Having everything from the problem details to the procedure in one place. "
" Oh, you got ads? Well, we need to make money too! But, you can use our platform to help you win SIH and take home the prize money. "
Get in touch if you need help bringing your ideas to life, or simply have an interesting idea you'd like to talk through. I'm always open to a conversation and exploring new possibilities together.