# GapexAI

> GapexAI is an AI document gap-analysis product by Codevally. Upload two versions of the same structured document (policy, contract, standard, SOP, specification) and get an evidence-backed report of every change: what is unchanged, what changed, what is new and what was removed.

## What it does
- Splits both documents (DOCX or PDF) into clean sections and filters out noise such as front matter, heading-only stubs and informative annexes.
- Matches every new section to its old counterpart by meaning, using local sentence embeddings (all-MiniLM-L6-v2, 384 dimensions) and cosine similarity.
- Uses a deterministic five-stage matcher: exact ID (with semantic veto), exact heading, renumbered-section inference, controlled split-section reuse, and greedy semantic matching with a 0.65 similarity floor.
- Resolves in-text cross-references one hop deep for both versions.
- Uses a large language model on matched section pairs only (never whole documents) to write two-sided findings that quote both versions. Section IDs come from the parser, never the model. Output is schema-validated, cost-capped and cached.
- Classifies each row as Full (unchanged or reworded), Part (materially changed) or Not (new or removed). Guarantees 100% coverage: every new section and every removed section gets a row.
- Flags figures, tables and images as "not analysed" so reviewers know what still needs a human check.
- Exports to Word, Excel, PDF, CSV and JSON. Includes a web dashboard with job history, analytics and multi-user admin.

## Accuracy
- Up to 91% agreement with expert manual reviewers on a real benchmark document pair.

## Use cases
Policy and SOP updates, contract revisions, standards and regulation updates, security frameworks (ISO 27001, SOC 2), RFP and tender responses, technical specifications, insurance wordings, quality documentation.

## Links
- Home: /
- Dashboard (sign-in required): /dashboard