All cases
AI document search implementation // technical corpus

AI search over technical documents: grounded RAG with page citations

The AI Geologist helps specialists search geological reports and technical references. It returns the source file, page and exact value or condition, searches by terms and meaning, and validates citations on the server. The approach also applies to other document collections.

Index
7,465 searchable chunks
Documents
204 files and 126 source records
Retrieval check
Expected document in the top 10 for 21 of 23 questions

What the system returns

A user asks a plain-language question and receives an answer with a document title, page, sheet or slide. Complex questions are split into multiple retrieval topics. Missing evidence for one part produces a clarification or refusal. A client report uploaded to the system does not automatically become a normative source.

Document pipeline

  1. 01Download the file and preserve its source URL, MIME type, size and SHA-256.
  2. 02Extract PDF, Word, Excel or PowerPoint structure; run OCR only for weak pages.
  3. 03Create chunks with page, sheet, cell range or slide addresses.
  4. 04Build lexical and semantic indexes.
  5. 05Generate only from retrieved evidence and validate citations on the server.
Workflow diagramGERON LABS

Trace the answer back to its document

  1. Files → source fragments

    Keep the document and page, sheet or slide address with each fragment.

  2. Exact + semantic retrieval

    Find identifiers by terms and paraphrases through embeddings.

  3. Answer with known references

    The model can select only fragments supplied in its context.

  4. Server-side citation checks

    Unknown references and unsupported figures block the answer. A specialist reviews critical conclusions.

RAG workflow from the geology project. No private reports or customer documents are shown.

Why retrieval uses two channels

FTS5/BM25 handles order numbers, references and exact terms. Embeddings retrieve paraphrases. Reciprocal Rank Fusion and metadata filters combine both rankings.

A complex question becomes two to four independent subqueries. Missing evidence for one step blocks a complete answer.

exact terms → FTS5 / BM25
plain-language question → embeddings
rankings → RRF → evidence set
claims → citation validation → answer

Preventing invented citations

The model may select only known chunk IDs. The server verifies that each ID exists, belongs to the stated document and contains the quoted text. An unknown citation, an uncited claim or a changed number forces a refusal.

Document text is treated as untrusted data. An instruction embedded in a PDF cannot change system rules or request a secret.

Numbers and document currency

A narrow water-withdrawal calculation runs in deterministic Decimal code. The model identifies explicit inputs, while the formula, substitution and result remain auditable.

A document being present in an official catalog does not prove that the edition is currently valid. That requires a separate status registry and verification source. A domain specialist still confirms critical conclusions.

Sources and links

SYS: COOKIE

We use technical browser storage so the site works correctly. Privacy Policy