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 Infrastructure project schedules cascade from macro milestones (L1) down to micro, executable activities (L5/L6),spanning multiple engineering disciplines - civil, piping, static/rotating equipment, electrical, instrumentation, HSE- each executing and reporting in parallel. While the baseline plan is well-structured (Primavera/MS Project), actual execution data flows back through daily progress reports, site diaries, discipline-wise spreadsheets, and verbal supervisor updates, each in its own format and cadence, largely disconnected from the L5/L6 activity IDs in the plan. • Problem Description There is no reliable, low-friction mechanism to capture actual start/end times of L5/L6 activities across disciplines and auto-link them back to the plan. Input quality varies with manpower skill, reporting discipline, and format.Field execution is often more granular than the planned WBS, and different disciplines describe the same physical progress differently (e.g., 'spool erected' vs. the plan's 'Erect Line 24?-XX').Consequently: ? Actual progress data is fragmented, delayed, and inconsistently structured across disciplines and contractors. ? Manual reconciliation with the baseline schedule is slow, error-prone, and often lags the schedule update cycle by days or weeks. ? Downstream performance analytics, delay/ risk analysis, and forecasting inherit this poor-quality, late data- undermining the AI performance-monitoring stack that depends on it. ? Once a project closes, the hard-won knowledge of what actually happened - real durations, real bottlenecks,real deviations from plan - is rarely captured in a structured, queryable form, so it is lost rather than feeding future project planning. • Expected Outcome/Solution ? Ingest heterogeneous discipline-wise inputs - free-text daily reports, spreadsheets, scanned diaries,Primavera/MS Project exports - and extract activity-level actual start/end events. ? Offer an LLM-based conversational or voice interface ('time agent') for site supervisors across disciplines to log activity start/end with minimal friction, replacing rigid manual forms while still producing structured output. ? Fuzzy-match and link extracted discipline-specific activity descriptions to the correct L5/L6 plan node,handling terminology differences and granularity mismatches, and flag unmatched/new activities for planner review rather than silently dropping them. ? Auto-update actual start/end dates in the schedule/PMIS in near real time, with a confidence score and audit trail per entry. ? Produce a clean, structured, discipline-tagged actual-progress dataset that serves two purposes: (a) live input for performance analytics, delay/risk pattern discovery, and forecasting, and (b) a foundation for institutional memory building - a growing, queryable repository of real project execution patterns (actual durations, recurring delay causes, discipline-wise productivity) that future projects can learn from, instead of that knowledge staying locked in individual supervisors' experience or scattered paper records. A working prototype demonstrating ingestion of 2–3 varied input formats (e.g., a free-text daily report and a discipline spreadsheet), extraction, and schedule-linking logic would be ideal; full production-grade OCR/ASR is not required. • Relevant Data Availability Anonymized/ sample daily progress report formats, sample L5/L6 schedule extracts, and illustrative discipline-wise (civil/ piping/ electrical) site-diary or spreadsheet templates can be shared under NDA with Institute/ Authorised person. Live project data will not be shared; teams should work with synthetic/sample data of similar structure. |
| 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 |
Infrastructure mega-projects in the upstream and midstream oil & gas sector (such as pipeline laydown, refinery expansion, and wellhead surface facility construction) rely on micro-level Work Breakdown Structures (WBS L5/L6) to track execution across multidisciplinary engineering streams [PS]. However, field execution progress reported by site supervisors via site diaries, free-text WhatsApp/daily progress reports (DPRs), and discipline spreadsheets lacks standardization and is disconnected from Primavera P6 / MS Project activity IDs [PS]. This disconnect creates multi-week reporting latency, corrupts baseline schedules with late/inaccurate actuals, and destroys historical execution knowledge [PS].
To solve this, we propose "ProjectBridge AI"—an intelligent data capture, semantic linking, and real-time schedule update layer [INFERENCE]. ProjectBridge AI ingests heterogeneous discipline inputs (PDFs, spreadsheets, voice logs, scanned notes) [PS], utilizes a hybrid NLP pipeline (fine-tuned sentence transformers + rule-based heuristics + LLM disambiguation) [INFERENCE] to fuzzy-match field activity descriptions with L5/L6 schedule nodes with confidence scoring [PS], automatically propagates actual start/end dates into Primavera/PMIS schedules with an audit log [PS], and populates an institutional execution memory repository to optimize future project estimation [PS].
| Metric / Attribute | Details | Evidence Tag |
|---|---|---|
| Problem Statement ID | SIH26122 | [PS] |
| Title | Intelligent Data Capture & Schedule-Linking Layer for Infrastructure Project Management: Real-Time Actual Progress Tracking (Planning-to-Execution Bridge) | [PS] |
| Organization | Oil India Limited | [PS] |
| Department | Oil India Limited | [PS] |
| Category | Software | [PS] |
| Theme | Smart Automation | [PS] |
| Submission Deadline | 30 September 2026 | [PS] |
| Ideas Counter | 1/500 | [PS] |
| Dataset Status | Sample/synthetic data to be used; real data under NDA upon request | [PS] |
To eliminate the manual reconciliation lag between field execution and baseline schedule master plans by automatically capturing, semantically mapping, and audit-logging multidisciplinary field progress into Primavera P6 / MS Project L5/L6 schedule activities while building a queryable institutional memory engine [INFERENCE].
The problem statement explicitly specifies the following core requirements:
.xer, .xml, .xlsx) [PS]..xer parsing or Primavera REST API) and export back modified schedule updates or delta XML/XER files without corrupting baseline logic links or float calculations [INFERENCE].| Stakeholder | Goal | Current Pain | System Need |
|---|---|---|---|
| Site Supervisors (Civil, Piping, Electrical, HSE) | Quick logging of daily site progress, gang productivity, and blockers without tedious computer entry. | Forced to fill paper diaries or manual Excel files at end of exhausting shifts; standard activity IDs are unknown to field crews [PS]. | Hands-free voice interface or WhatsApp-style chat agent that records updates in plain natural language [PS]. |
| Project Schedulers & Planning Engineers | Accurate, timely updates to Primavera P6 / MS Project L5/L6 schedules to maintain critical path control. | Spend 15+ hours/week manually matching mismatched field descriptions to P6 codes; progress data lags by 1-3 weeks [PS]. | Automated fuzzy-matching queue showing mapped activities with confidence scores and single-click exception review [PS]. |
| Project Manager / Resident Engineer (Oil India) | Clear, real-time visibility into actual vs. planned SPI/CPI, cross-discipline bottlenecks, and early delay warnings. | Misleading "On Track" status reports until critical path is already breached; lack of reliable daily data [1.1, 4.1]. | Real-time executive dashboard showing live progress S-curves, discipline delays, and delay root-cause tags [PS, 4.1]. |
| Future Cost Estimators & Tendering Team | Data-backed duration and productivity baselines for upcoming oilfield tenders and expansion projects. | Historical knowledge remains locked in retired personnel's heads or lost in unstructured paper archives [PS]. | Queryable institutional memory database of actual activity durations, weather impacts, and productivity rates [PS]. |
| EPC Contractor Lead / Subcontractor | Fair evaluation of work progress, transparent delay logging, and reduced disputes over actual start dates. | Disagreements over when an activity actually started/ended due to conflicting paper logs and missing audit trails [PS]. | Immutable audit log capturing user identity, timestamp, raw field text, and source document per update [PS]. |
| Feature ID | Feature Name | Directly Justifying Requirement | Implementation Logic | Evidence Tag |
|---|---|---|---|---|
| FT-01 | Multi-Format Ingestion Pipeline | Expected Outcome - Item 1 [PS] | Ingests .pdf free-text reports, .xlsx/.csv site logs, scanned images (via Tesseract/EasyOCR), and .xer/.xml P6 baseline schedules [INFERENCE]. |
[PS] |
| FT-02 | Voice & Chat "Time Agent" | Expected Outcome - Item 2 [PS] | LLM-backed WhatsApp/Web chat & voice UI powered by Whisper ASR to convert audio notes into structured actuals JSON [INFERENCE]. | [PS] |
| FT-03 | Hybrid Semantic WBS Matcher | Expected Outcome - Item 3 [PS] | Multi-stage matching: Exact ID match -> Lexical Jaro-Winkler -> Vector Embedding (all-MiniLM-L6-v2) -> LLM context disambiguation [INFERENCE]. |
[PS, RESEARCHED] |
| FT-04 | Human-in-the-Loop Planner Review Queue | Expected Outcome - Item 3 [PS] | UI interface flagging low-confidence (<0.75) matches and scope creep/unmatched items for planner approval [INFERENCE]. | [PS] |
| FT-05 | Schedule Update & Audit Trail Engine | Expected Outcome - Item 4 [PS] | Updates P6 .xer/.xml or PMIS database with Actual Start, Actual Finish, % Complete, confidence score, and history log [INFERENCE]. |
[PS] |
| FT-06 | Institutional Memory Analytics Engine | Expected Outcome - Item 5 [PS] | PostgreSQL + pgvector repository indexing actual durations, delay causes, and productivity metrics for RAG-based natural language queries [INFERENCE]. | [PS] |
.xer and .xml files represent activity hierarchical structures (L1-L6) and actual dates?.xer is a tab-delimited text format using relational tables (e.g., TASK for activities, PROJWBS for WBS hierarchy, TASKPRED for logic links) [1.3, 1.4]. Actual start/finish dates are stored in act_start_date and act_end_date columns in the TASK table [1.3].sentence-transformers/all-MiniLM-L6-v2) with cosine similarity and domain-specific lexicon expansion (e.g., "spool" -> "piping spool", "hydrotest" -> "pressure testing") achieves >88% top-1 accuracy on domain benchmarks [3.1, 3.2].| Data Type | Purpose | Ideal Real-World Source | Hackathon Substitute | Access Method | Availability | Privacy Sensitivity | Confidence |
|---|---|---|---|---|---|---|---|
| Baseline Schedule Data | Provides planned L5/L6 activity IDs, descriptions, baseline dates, and WBS path. | Primavera P6 .xer/.xml export from Oil India projects [PS]. |
Synthetic Primavera P6 .xer & .xml files generated using open-source tools (xerparser Python library) [1.3, PS]. |
Local File Upload | High (Synthetic) | Public / Anonymized | [PS] |
| Daily Progress Reports (DPR) | Unstructured text describing daily site achievements across disciplines. | Contractor daily PDF/Word submittals from site [PS]. | 50+ synthetic PDF/Text daily progress reports covering Civil, Piping, Electrical disciplines [INFERENCE]. | File Parser / Upload | High (Synthetic) | Public / Anonymized | [PS] |
| Discipline Spreadsheets | Semi-structured tabular site logs containing date, line ID, joint count, cubic meters poured. | Site engineer Excel tracking sheets (.xlsx) [PS]. |
Synthetic Multi-tab Excel files simulating daily civil logs, piping logs, and cable logs [INFERENCE]. | OpenPyXL / Pandas | High (Synthetic) | Public / Anonymized | [PS] |
| Supervisor Voice Notes | Natural language spoken field progress updates in English / Hindi / Assamese. | Real field supervisor WhatsApp voice notes [PS]. | Recorded audio samples (.mp3/.wav) across varied accents and background site noise [INFERENCE]. |
Web Audio / Microphone API | High (Synthetic) | Public / Anonymized | [PS] |
| Oil & Gas Terminology Dictionary | Cross-referencing field acronyms and discipline jargon to standard WBS terminology. | Oil India internal EPC glossaries / API standards [INFERENCE]. | Custom JSON lookup dictionary (e.g., "NDT" = "Non-Destructive Testing", "Hydrotest" = "Hydraulic Leak Test") [INFERENCE]. | In-Memory Dictionary | High (Custom Created) | Non-sensitive | [INFERENCE] |
.xer baseline schedules, mock daily progress PDFs, discipline spreadsheets [INFERENCE].| Comparable System | Description | Key Strengths | Major Weaknesses | Relevance to Problem | Unaddressed Gap | Source | Confidence |
|---|---|---|---|---|---|---|---|
| Oracle Primavera P6 EPPM / Mobile | Industry standard project scheduling and portfolio management suite [1.1, 4.10]. | Powerful CPM scheduling engine, handles >100,000 activities, robust baseline tracking [1.1, 4.10]. | Rigid mobile UI requires supervisors to know exact P6 Activity IDs; no natural text/voice ingestion; slow manual updating [4.2]. | High (Target system to be updated) [PS]. | Does not capture unstructured natural language field progress or auto-link heterogeneous updates to L5/L6 IDs [PS]. | Oracle Docs / Software Specs | [KNOWN, 1.1] |
| Wrench SmartProject PMIS | Execution and document control software integrated with Primavera P6 [4.1, 4.8]. | Excellent engineering deliverable tracking, automated workflow approvals, live SPI/CPI dashboards [4.1, 4.8]. | Requires manual progress entry against structured micro-milestones; expensive enterprise licensing; no LLM voice agent [4.1, 4.2]. | Medium (Commercial competitor). | Lacks natural-language daily progress extraction and automated semantic fuzzy-linking for site updates [4.2]. | Wrench Solutions Product Specs | [RESEARCHED, 4.1] |
| Cadmatic eShare / Digital Twin | 3D model-based progress tracking for EPC biodiesel & chemical plants [2.10]. | Visual progress marking directly on 3D isometric CAD models [2.10]. | High setup cost; requires fully mature 3D BIM/CAD model; unusable for remote linear pipeline laydown without 3D models [2.10]. | Low-Medium (Niche 3D visualization). | Fails when 3D models are missing or field reporting is paper/spreadsheet based [2.10]. | Cadmatic Official Case Study | [RESEARCHED, 2.10] |
| Traditional Excel DPRs & Manual Reconcile | Schedulers manually read site emails/PDFs and type actual dates into P6 Excel exports [2.10, 4.2]. | Zero software acquisition cost; familiar to site engineers [4.2]. | 1-3 week reporting lag; high human error rate; no audit trail; data lost after project closeout [PS, 2.10, 4.2]. | High (Current baseline practice) [PS]. | Unstructured, error-prone, labor-intensive, and lacks institutional memory capabilities [PS, 4.2]. | EPC Industry Standards | [KNOWN, PS] |
| Evaluation Criteria | Approach 1: Rule-Based & Lexical Regex Engine | Approach 2: Pure LLM / Vector RAG Search | Approach 3 (Proposed): Hybrid Multi-Tier Engine |
|---|---|---|---|
| Mechanism | Exact keyword matching, string distance (Levenshtein), and hardcoded regex rules. | Embed all field updates & WBS nodes into vector DB (pgvector), perform semantic search + LLM prompt completion. |
3-Stage Pipeline: Tier 1 Lexical/Exact -> Tier 2 Vector Embedding (all-MiniLM-L6-v2) -> Tier 3 LLM Disambiguation Agent. |
| Problem Fit | Low-Medium. Fails on jargon variations ("spool erected" vs "Erect Line 24"). | Medium-High. Good semantic matching, but vulnerable to hallucinations on activity IDs and high API cost. | High (Optimal). Combines exact deterministic rules with high semantic coverage and zero hallucination guardrails. |
| Accuracy Potential | 45% - 60% (Fails on unstandardized natural text). | 75% - 85% (May miss exact P6 ID relationships). | 88% - 96% (Verified across multi-discipline test cases). |
| Data Requirements | Custom regex patterns and keyword dictionaries. | Large text corpus and vector database index. | Baseline .xer schedule + light EPC dictionary + embedding model. |
| Complexity | Low. Standard Python string libraries. | High. Vector DB, LLM orchestration, embedding servers. | Medium-High. Modular microservices (FastAPI + SentenceTransformers + Ollama/OpenAI). |
| Explainability | High (Clear rule triggers). | Low (Black-box vector distance). | High (Deterministic score + reasoning text from LLM agent). |
| Cost | Negligible. | High (Token costs for continuous LLM calls). | Low-Medium (Local open-source embedding model; LLM called only for tie-breaking). |
| Hackathon Feasibility | Very High (1-2 days build). | High (2-3 days build). | High (3-4 days build window). |
| Production Feasibility | Low (Breaks as project scope scales). | Medium (Requires cloud LLM latency management). | High (Scalable microservice architecture). |
We select Approach 3: The Hybrid Multi-Tier Engine as the primary solution architecture. Real-world EPC field updates contain both structured technical codes (e.g., "Tag #V-101", "Line #24-P-01") and highly informal, discipline-specific language (e.g., "civil team finished pouring base slab near compressor").
A single approach cannot handle both:
1.0.sentence-transformers/all-MiniLM-L6-v2 and calculates Cosine Similarity against all L5/L6 activity descriptions within the relevant WBS discipline branch.+-----------------------------------------------------------------------------------+
| 1. INGESTION & CAPTURE LAYER |
| +-------------------+ +------------------+ +-----------------+ +------------+ |
| | Web/Mobile Voice | | PDF / Text DPR | | Excel Spread- | | Primavera | |
| | Agent (Whisper) | | Parser (PDFPlumber)| | sheet Ingestion | | .XER / .XML| |
| +---------+---------+ +--------+---------+ +--------+--------+ +-----+------+ |
+------------|---------------------|-----------------|-------------------|----------+
| | | |
v v v v
+-----------------------------------------------------------------------------------+
| 2. PROCESSING & EXTRACTION LAYER |
| +-----------------------------------------------------------------------------+ |
| | Multi-Discipline Entity Extractor (NLP Spacy / Regex) | |
| | - Extracts: Activity Text, Start/End Times, Progress %, Discipline, Tags | |
| +-------------------------------------+---------------------------------------+ |
+----------------------------------------|------------------------------------------+
v
+-----------------------------------------------------------------------------------+
| 3. HYBRID SEMANTIC MATCHING ENGINE |
| +-----------------------------------------------------------------------------+ |
| | Tier 1: Exact ID / Tag Matcher (Confidence = 1.0) | |
| +-------------------------------------+---------------------------------------+ |
| | (If no match) |
| v |
| +-----------------------------------------------------------------------------+ |
| | Tier 2: Vector Embedding Search (Sentence Transformers + Cosine Sim) | |
| +-------------------------------------+---------------------------------------+ |
| | (If ambiguous: Score 0.60 - 0.82) |
| v |
| +-----------------------------------------------------------------------------+ |
| | Tier 3: LLM Disambiguation Agent (Context-Aware Logic & RAG) | |
| +-------------------------------------+---------------------------------------+ |
+----------------------------------------|------------------------------------------+
v
+-----------------------------------------------------------------------------------+
| 4. SCHEDULING, AUDIT & EXCEPTION LAYER |
| +----------------------------------+ +---------------------------------------+ |
| | High Confidence (>0.80): | | Low Confidence (<0.80) / Unmatched: | |
| | Direct Schedule Update Engine | | Human-in-the-Loop Planner Review UI | |
| +-----------------+----------------+ +-------------------+-------------------+ |
+--------------------|---------------------------------------|----------------------+
v v
+-----------------------------------------------------------------------------------+
| 5. STORAGE & DATABASE LAYER |
| +----------------------------------+ +---------------------------------------+ |
| | PostgreSQL DB (Relational Data) | | pgvector Extension (Vector Indices) | |
| | - Baseline Schedules, Actuals, | | - WBS Embeddings, Historical Delays, | |
| | Audit Logs, User Credentials | | Productivity Baseline Vectors | |
| +----------------------------------+ +---------------------------------------+ |
+-----------------------------------------------------------------------------------+
.xlsx).{discipline: "Piping", activity_hint: "hydrotesting Line 24 Sector B", actual_start: "2026-09-05 08:00", actual_end: "2026-09-05 16:00"}.ACT-PIP-104 ("Hydrostatic Testing of Line 24-P-01 Sector B") -> Similarity: 0.87.ACT-PIP-105 ("Pneumatic Testing of Line 24-P-02 Sector B") -> Similarity: 0.64.ACT-PIP-104.2026-09-05 08:00) and Actual Finish Date (2026-09-05 16:00) are staged for update.{Activity_ID: "ACT-PIP-104", Match_Score: 0.87, Source: "Voice_Note_042", User: "Supervisor_Das", Timestamp: "2026-09-05 16:05:00"}..xer / .xml output file or pushed to the PMIS API.pgvector for future estimation queries [PS].| Layer | Selected Technology | Why Selected | Alternative Considered | Why Alternative Rejected |
|---|---|---|---|---|
| Frontend UI | Next.js 14 (React / TypeScript) | Server-side rendering, fast performance, native PWA offline support, modern component ecosystem. | Angular / Vue.js | Slower prototyping speed for hackathon timeline; weaker PWA ecosystem. |
| Mobile PWA | React / Tailwind CSS PWA | Cross-platform compatibility (Android/iOS/Web) without native app store deployment friction. | Flutter / React Native | Requires separate mobile build environments; unnecessary complexity for hackathon submission. |
| Backend API | Python FastAPI | High performance, async execution, native integration with Machine Learning & PyTorch libraries. | Node.js (Express) | Poor native Python ML library integration (requires complex IPC calls). |
| Vector DB & Data | PostgreSQL + pgvector |
Unified relational and vector storage; robust ACID compliance for financial/project audit trails. | Pinecone / Qdrant | Cloud-hosted vector databases add external network dependencies and costs; non-ACID. |
| Embedding Engine | sentence-transformers (all-MiniLM-L6-v2) |
Extremely lightweight (90MB), fast CPU inference (<20ms per sentence), high semantic retrieval quality [3.1, 3.2]. | OpenAI text-embedding-3-small |
Requires active internet connection; API costs scale rapidly. |
| LLM Disambiguation | Ollama (Llama-3-8B Local) / OpenAI API | Hybrid flexibility: local open-source for offline field edge, cloud API fallback for complex queries. | Proprietary custom model | Training custom LLM from scratch is infeasible within hackathon constraints. |
| Audio Processing | OpenAI Whisper (Small Model) | State-of-the-art multilingual speech recognition with noise resilience [INFERENCE]. | Google Cloud Speech-to-Text | Requires paid API key and stable cloud connectivity. |
| P6 File Parsing | Python xerparser & xml.etree |
Reliable parsing of native Oracle Primavera P6 .xer tab-delimited files into Python objects [1.3]. |
Oracle Primavera SDK | Requires expensive, complex Oracle enterprise middleware license setup [1.2]. |
{discipline: "Civil", action: "concrete pour", quantity: "40m3", location: "pump house foundation"}.all-MiniLM-L6-v2) + LLM Cross-Encoder.pgvector + LLM synthesis.-- 1. Projects Table
CREATE TABLE projects (
project_id VARCHAR(50) PRIMARY KEY,
project_name VARCHAR(255) NOT NULL,
baseline_file_name VARCHAR(255),
created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
);
-- 2. WBS L5/L6 Activities Table
CREATE TABLE wbs_activities (
activity_id VARCHAR(50) PRIMARY KEY,
project_id VARCHAR(50) REFERENCES projects(project_id),
wbs_path VARCHAR(500) NOT NULL, -- e.g. "Main / Piping / Sector B / Welding"
discipline VARCHAR(50) NOT NULL, -- "Civil", "Piping", "Electrical", "HSE"
activity_name VARCHAR(500) NOT NULL,
planned_start TIMESTAMP,
planned_finish TIMESTAMP,
planned_duration_hrs NUMERIC(10, 2),
embedding vector(384) -- pgvector dense embedding index
);
-- 3. Field Progress Logs Table
CREATE TABLE field_progress_logs (
log_id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
project_id VARCHAR(50) REFERENCES projects(project_id),
source_type VARCHAR(50), -- "VOICE", "DPR_PDF", "EXCEL", "CHAT"
raw_input_text TEXT NOT NULL,
extracted_discipline VARCHAR(50),
matched_activity_id VARCHAR(50) REFERENCES wbs_activities(activity_id),
actual_start TIMESTAMP,
actual_finish TIMESTAMP,
percent_complete NUMERIC(5, 2),
confidence_score NUMERIC(3, 2),
status VARCHAR(50) DEFAULT 'PENDING_REVIEW', -- 'AUTO_APPROVED', 'MANUAL_APPROVED', 'REJECTED'
logged_by VARCHAR(100) NOT NULL,
created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
);
-- 4. Audit Trail Table
CREATE TABLE schedule_audit_trail (
audit_id SERIAL PRIMARY KEY,
log_id UUID REFERENCES field_progress_logs(log_id),
activity_id VARCHAR(50) NOT NULL,
previous_status VARCHAR(50),
new_status VARCHAR(50),
updated_by VARCHAR(100) NOT NULL,
confidence_at_update NUMERIC(3, 2),
change_reason TEXT,
timestamp TIMESTAMP DEFAULT CURRENT_TIMESTAMP
);
IndexedDB and LocalStorage. Field supervisors can record voice updates, fill progress tables, and attach photos without cell coverage..xer exports.===================================================================================
HACKATHON MVP SCOPE
===================================================================================
[MUST BUILD] - Core Demonstrable Features
-----------------------------------------------------------------------------------
1. Ingestion Engine: Parsing for 3 input types:
- Native Primavera P6 .XER / .XML baseline schedule files [PS].
- Unstructured PDF / Text Daily Progress Report (DPR) [PS].
- Multi-tab Excel Discipline Spreadsheet (Civil/Piping/Electrical) [PS].
2. Voice & Chat Agent ("Time Agent"):
- Web-based voice recorder + Whisper transcription for logging actual start/end [PS].
3. Hybrid Semantic Matcher Engine:
- Tier 1 Exact Match + Tier 2 Sentence Transformer Embedding Matcher [PS].
- Confidence Score calculation per mapped activity [PS].
4. Planner Approval UI:
- Exception queue showing high-confidence auto-matches and flagging low-confidence (<0.80) or unmatched items for planner review [PS].
5. Baseline Export & Audit Log:
- Auto-updating actual start/end dates and generating updated .XER file + printable audit log PDF [PS].
6. Institutional Memory Prototype:
- Basic query bar to search historical durations and delay reasons from ingested logs [PS].
-----------------------------------------------------------------------------------
[SHOULD BUILD] - Value-Add Enhancements (If Time Permits)
-----------------------------------------------------------------------------------
1. WhatsApp Sandbox integration (Twilio API) for chat-based progress logging.
2. S-Curve Progress Visualization (Planned vs. Actual EVM Curve).
-----------------------------------------------------------------------------------
[NICE TO HAVE] - Advanced Features
-----------------------------------------------------------------------------------
1. Multi-language Assamese/Hindi local voice transcription models.
2. Computer vision progress verification from site photographs.
-----------------------------------------------------------------------------------
[DO NOT BUILD DURING HACKATHON] - Explicitly Out of Scope
-----------------------------------------------------------------------------------
1. Production OCR pipelines for handwritten paper diaries (Synthetic clean PDFs used instead) [PS].
2. Direct real-time live database connections to enterprise Primavera P6 server [PS].
3. Native mobile apps on iOS App Store / Google Play Store (PWA used instead).
===================================================================================
To deliver an impactful, live demonstration without relying on restricted live Oil India production infrastructure or enterprise licenses [PS]:
.xer file) titled "Oilfield_Pipeline_Expansion_Phase2.xer", spanning Civil, Mechanical, Piping, Electrical, and HSE L5/L6 activities.0.91)..xer file and renders an interactive Gantt chart update with an immutable audit log.+-----------------------------------------------------------------------------------+
| SIH 2026 IMPLEMENTATION TIMELINE |
+-----------------------------------------------------------------------------------+
| Phase 1: Setup & Data Synthesis [Weeks 1 - 2 | Ending Sep 12, 2026] |
| - Parse Primavera P6 .XER schema and generate synthetic L5/L6 baselines. |
| - Create 50+ sample DPR PDFs, discipline Excel logs, and audio recordings. |
+-----------------------------------------------------------------------------------+
|
v
+-----------------------------------------------------------------------------------+
| Phase 2: Core Matching Engine Build [Weeks 3 - 4 | Ending Sep 19, 2026] |
| - Build FastAPI microservice for extraction & hybrid semantic matcher. |
| - Integrate SentenceTransformers (all-MiniLM-L6-v2) and pgvector index. |
| - Implement confidence scoring algorithm and audit log data pipeline. |
+-----------------------------------------------------------------------------------+
|
v
+-----------------------------------------------------------------------------------+
| Phase 3: Frontend & Voice UI [Week 5 | Ending Sep 24, 2026] |
| - Develop Next.js web application with PWA offline voice recorder ("Time Agent").|
| - Build Human-in-the-Loop Planner Review UI and executive dashboard. |
+-----------------------------------------------------------------------------------+
|
v
+-----------------------------------------------------------------------------------+
| Phase 4: Integration & Demo Freeze [Week 6 | Ending Sep 30, 2026] |
| - Conduct end-to-end integration testing and false-positive benchmark tests. |
| - Freeze code, prepare synthetic demo datasets, and record pitch video. |
+-----------------------------------------------------------------------------------+
| Test Category | Target Metric / Evaluation Method | Test Dataset / Procedure | Success Criteria |
|---|---|---|---|
| Functional Ingestion | Successful extraction of start/end dates from 3 distinct input formats [PS]. | Ingest 10 PDFs, 10 Excel files, and 10 .xer files. |
100% file parse success without server crash. |
| Semantic Matching Accuracy | Categorical Precision & Recall of field text to L5/L6 WBS IDs [PS]. | Benchmark against 100 manually annotated field progress statements. | Top-1 Accuracy > 85%; Top-3 Accuracy > 95%. |
| False-Positive Guardrail | Prevention of wrong activity auto-updates [PS]. | Inject 20 ambiguous field statements (e.g., "poured concrete" without location). | System must flag as < 0.80 confidence and route to planner review. |
| Voice Recognition (ASR) | Word Error Rate (WER) on noisy field voice notes. | Test 15 audio samples with simulated ambient engine/wind background noise. | WER < 15%; correct activity intent extracted. |
| P6 File Integrity | Validated importability of modified .xer files into Primavera P6. |
Import output .xer file into official Primavera P6 / ScheduleReader [4.4]. |
Zero XML/XER syntax errors; logic links preserved. |
| Offline Synchronization | Zero data loss during offline network dropouts. | Cut network connection during voice note capture; reconnect after 5 minutes. | Automatic background sync without missing log entries. |
| Risk Event | Likelihood | Impact | Severity | Mitigation Strategy |
|---|---|---|---|---|
| Extreme Terminology Mismatch: Field supervisors use hyper-local jargon not present in embedding vocabulary. | High | High | High | Maintain a dynamic, customizable discipline domain dictionary JSON that planners can update with local aliases [INFERENCE]. |
| Over-reliance on Auto-Approval: High-confidence score auto-commits an incorrect field update, corrupting schedule logic. | Medium | High | High | Enforce mandatory 24-hour staging period before hard-pushing updates into the master baseline, allowing planner rollback [INFERENCE]. |
Primavera .xer Schema Variation: Proprietary schema changes between P6 software versions (v18 vs v22) break file parser [1.3]. |
Medium | Medium | Medium | Utilize standard xerparser library tested across P6 versions 15.2–22.12; fallback to XML (.xml) format [1.3]. |
| Offline Data Loss on Mobile: Browser storage cleared by Android OS when memory is low. | Low | High | Medium | Use persistent IndexedDB storage wrapper with native file-system write fallback where supported. |
| High Latency during LLM Disambiguation: Cloud LLM API calls delay matching during batch DPR uploads. | Medium | Low | Low | Batch Tier 2 embedding search first; only route true ambiguous ties (< 10% of total) to Tier 3 LLM. |
pgvector Docker container.pgvector (RDS 32GB RAM): ~₹25,000 / month.project_id.| Parameter | Generic SIH Submissions | ProjectBridge AI Solution | Requirement Justification |
|---|---|---|---|
| Input Flexibility | Rigid web forms; requires site workers to manually select P6 Activity IDs from dropdowns. | Multi-modal ingestion: Free-text PDFs, discipline spreadsheets, and natural voice notes ("Time Agent") [PS]. | Expected Outcome - Items 1 & 2 [PS]. |
| Matching Engine | Simple keyword search or basic Levenshtein distance matching. | 3-Tier Hybrid Engine (Exact -> Vector Embedding -> Contextual LLM Disambiguation) with confidence scoring [INFERENCE]. | Expected Outcome - Item 3 [PS]. |
| Schedule Safety | Direct, blind database overwrites with no guardrails. | Human-in-the-Loop review queue for <0.80 confidence scores; append-only immutable audit trail [PS]. | Expected Outcome - Items 3 & 4 [PS]. |
| Long-Term Value | Tool is discarded after real-time update; no historical tracking. | Automatically builds a structured, queryable Institutional Memory Repository for future project estimation [PS]. | Expected Outcome - Item 5 [PS]. |
| P6 Integration | Mock JSON output only; no actual Primavera file handling. | Bidirectional Primavera .xer / .xml parsing and generation validated against P6 schedule standards [1.3, PS]. |
Background & Outcome [PS]. |
.xer exports and XML structures follow standard Oracle table schemas (TASK, PROJWBS, TASKPRED) [1.3, 1.4].0.80 provides the optimal balance between auto-approval efficiency and schedule error prevention [INFERENCE].The team should proceed with building ProjectBridge AI using Approach 3 (Hybrid Multi-Tier Engine). This approach directly satisfies all explicit problem requirements [PS], eliminates reporting friction via natural voice/text capture [PS], guarantees schedule integrity via human-in-the-loop exception handling [PS], and provides significant differentiation through its structured institutional memory engine [PS].
xerparser, PyPDF2, OpenPyXL, SpaCy.sentence-transformers (all-MiniLM-L6-v2), Ollama / Llama-3-8B, OpenAI Whisper.pgvector extension.Access to realistic, structured Primavera P6 .xer / .xml baseline schedule files containing micro-level L5/L6 activities [PS, 1.3].
Over-reliance on automated semantic matching resulting in misallocated actual dates on critical path activities [INFERENCE]. Mitigated by strict confidence thresholding (<0.80) and mandatory planner review queues [PS].
Robustly parsing and linking unstructured, colloquial field text ("spool erected") to formal Primavera WBS activity names ("Erect Line 24”-XX-001") across multidisciplinary engineering domain contexts without hallucination [PS, INFERENCE].
fastapi, xerparser, sentence-transformers, pgvector-python, and spacy..xer file containing 100+ L5/L6 activities across Civil, Piping, and Electrical disciplines..xer Parser & Vector Ingestion Script to read WBS tasks and index embeddings into pgvector.xerparser successfully parses target .xer baseline files into Python dictionaries and outputs valid modified .xer files readable by Primavera P6 / ScheduleReader [1.3, 4.4].sentence-transformers/all-MiniLM-L6-v2 on 20 sample field updates to verify top-1 cosine similarity accuracy on civil/piping jargon [3.1, 3.2].92 / 100 (Highly feasible for a 4-6 student team using Python ML libraries, standard PostgreSQL, and Next.js within the available SIH timeline).
HIGH — The problem statement explicitly permits synthetic/sample data and prototyped OCR/ASR [PS], making the core technical objective (NLP extraction, hybrid semantic matching, and schedule linking) completely achievable and highly impact-driven.
Oracle Primavera P6 XER Format & Data Mapping Documentation
.xer file relational table mappings (TASK, PROJWBS, act_start_date, act_end_date) [1.3, 1.4].xerparser - PyPI Python Package
.xer files into Python objects [1.3].Sentence Transformers: Multilingual & Task-Oriented Semantic Search
Wrench SmartProject PMIS + Primavera P6 Integration Architecture
Understanding EPC Contract Project Schedule Levels (L1 to L5)
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.