How BhoomiScan verifies Odisha land documents
Step 1. Upload the three core documents
Drag in the Encumbrance Certificate (EC, Form 25), the Record of Rights (ROR / Khatiyan — sabak and hal both work), and the Sale Deed. Extra papers — mutation order, partition deed, NOC, conversion order — go into the supporting-documents tray. Slots are rigid: BhoomiScan does not auto-classify, so you stay in control of which file is which.
Step 2. OCR with Odia language support
Each PDF is run through Reducto L1 OCR. Multi-page ECs (long 30-year searches can stretch to a few hundred pages) are chunked and processed in parallel. Odia and English rows are both extracted, including stamped seals, marginal notes, and tabular khatiyan layouts that traditional PDF text extractors miss.
Step 3. Per-document field extraction
A document-type-specific parser (EC, ROR, Sale Deed) pulls structured fields: khata, plot, area, kissam, tenant names, transaction entries, encumbrance rows, registration date, SRO, parties, sale consideration. Deterministic regex covers the high-confidence path; an LLM-assisted pass handles ambiguous Odia text and edge layouts.
Step 4. Chain reasoner correlates the three sources
Gemini 2.5 Pro takes the extracted fields plus the first three pages of each source for context (8 KB budget) and reasons across documents. It cross-references whether the seller in the Sale Deed matches the tenant in the latest ROR, whether the EC reflects the registered Sale Deed, whether mutation is current, and whether identity / transaction details reconcile.
Step 5. Five chain checks, colour-coded
The unified report surfaces five findings: ownership continuity, encumbrance posture, mutation status, identity reconciliation, and transaction completeness. Each is marked Critical, Resolvable, or Clean. Every flag links back to the source page and paragraph so you can verify the underlying record yourself before signing the legal opinion.
Frequently asked questions
Does the pipeline read Odia text in khatiyan PDFs?
Yes. Reducto L1 plus Odia-specific post-processing handles both Odia and English rows in the Bhulekh khatiyan layout. Sabak and hal formats are both supported.
What if a page in the EC is rotated or low-resolution?
OCR has retry-with-backoff. Pages that fail the first pass are re-tried via a vision-LLM fallback so a single bad scan does not break the run.
Can I upload more than the three core documents?
Yes — the supporting-documents tray accepts mutation orders, partition deeds, conversion orders, and any NOC.
Where does the verification report come from — is it cached?
Each scan is a fresh run. The chain reasoner is non-cached: it reads the actual uploaded files for that specific case.