AI-Powered Geological, Mining and other Reporting Solution for CMPDI/CIL subsidiaries
Ministry of Coal · Miscellaneous · Software
Genuinely low-risk with real public documents and a reliable demo, but everyone is building document chatbots this year, so win on traceable numeric extraction with citations rather than on the conversation.
What it actually is
When Parliament asks Coal India a question, staff have to dig production figures out of scanned PDFs, spreadsheets and old archives by hand, which is slow and error-prone. The ask is a system that reads all those documents automatically and can answer questions from them. It should also generate the standard reports without someone assembling them each time.
What to build
A document intelligence layer over the corpus types the statement names — scanned PDFs, digital documents, spreadsheets, images and historical archives — with OCR and table extraction pulling production and geological figures into a structured store that keeps a traceable link from every number back to its source page, an automated report generator that assembles the recurring report formats from that store, a topic identification and word cloud module over the corpus, and a natural language query interface that answers a parliamentary-style question with the figure, the reporting period and a citation to the exact document and page it came from.
Smallest thing that wins the room
Ask a production question in plain language, get a specific tonnage figure back, and click through the citation to the scanned page it was lifted from with the table cell highlighted.
How crowded this one gets
A guess, projected from the 2025 statements — the last year where both the submission counts and the winners were published.
Quieter than 28% of the 226 · #163 of 226 by expected field
A normal-sized field. Your idea has to be good, not miraculous.
Why: central ministry statements sat below the average.
This is a guess, not a fact
Nobody has published 2026’s numbers yet. This is an analysed estimate from last year’s pattern, so please do not take it as the truth — check the live counter on the SIH portal before you decide anything. The range covers the middle half of likely outcomes, so one statement in two lands outside it. Entry closes at 500 ideas per statement, so no range goes past that — a statement that reaches the cap fills and shuts rather than drawing an unlimited crowd. The model reads only three things a team can see before choosing — software or hardware, the theme, and what kind of body posted it — and those explain about a quarter of the variation in last year’s field sizes (R² 0.25 on held-out statements). Trust the band more than the number, and the ordering more than either. It cannot see how good your idea is, which is the part that actually decides it.
The scores
The number is the shorthand. The line under it is the reason.
Acceptance potential
3/5Real public documents, a tractable scope and a demo that works reliably are all in its favour, but document question-answering is now among the most submitted shapes at any hackathon, so your standing depends entirely on doing traceable numeric extraction properly rather than shipping another chat interface.
Feasibility
4/5Coal India and CMPDI publish annual reports, production statistics and provisional figures openly, so you can assemble a genuine corpus of exactly the document types the statement describes, and OCR, table extraction and grounded retrieval over PDFs are all mature and well-supported.
Innovation scope
3/5Three modules are named but nothing about their implementation is prescribed, so the extraction strategy, the grounding mechanism and the report generation approach are open, though the overall shape as a document-query system is fixed.
Clarity
3/5The three desired outcomes are listed but the statement is loose around the edges — 'any other production figures' leaves the corpus undefined, a word cloud module sits oddly beside a query system, and the benefit metrics are written as unbounded aspirations rather than targets.
Effort
MediumThe core is an ingestion pipeline, a structured store and a grounded query interface, all of which are well-trodden with strong libraries; the only genuinely fiddly part is reliable table extraction from poor scans, and the statement asks for three modules rather than a dozen.
Demo-ability
EasyAsking a question in plain English and clicking through to the highlighted table cell in a scanned annual report is instantly convincing and requires no domain knowledge from the judge.
In its favour
- Green flag: Coal India and CMPDI publish exactly the document types described — annual reports, production statistics, geological summaries — so your corpus is real and downloadable rather than mocked
- Green flag: Numeric figures are objectively checkable, which means you can report extraction accuracy against a hand-verified sample and make correctness rather than fluency your claim
- Green flag: Filed under Miscellaneous, so it will not appear for anyone filtering by a substantive theme
- Green flag: The scope is unusually contained for a ministry statement — three modules, not fifteen — which makes finishing properly realistic
Against it
- Red flag: Document question-answering is now an extremely common submission and a generic chat-over-PDF build will be indistinguishable from many others in the room
- Red flag: The real difficulty is table extraction from degraded scans where a misread digit silently produces a wrong tonnage, and a system that answers confidently with a wrong number is worse than one that declines
- Red flag: Parliamentary questions demand exactness and attribution, so any answer without a verifiable citation to source page defeats the purpose — grounding is the requirement, not a nice-to-have
- Red flag: A word cloud module is asked for alongside a query system and it adds nothing analytically; build it because it is specified, but do not let it near your demo
What you will be writing
- Camelot / Tabula + PaddleOCR table extraction
- hybrid retrieval with span-level citation grounding
- DuckDB structured figure store
- BERTopic topic identification
- Jinja-templated report generation
- FastAPI + React query interface
- Document intelligence
- Coal sector reporting
- Government information systems
Prior art to read before you start
question answering over scanned reports · table extraction from PDF archives · automated report generation
Analysed by Claude Opus. Every score above is a judgment call with its reasoning attached — kindly cross-check this against the official statement on the SIH portal before your team commits to it.