Why I Built VaultDocs AI: Local-First Contract Risk Screening with Open Models

Description and use

VaultDocs AI is a local-first contract risk screening tool that uses open-weight LLMs via Ollama, parses PDFs, retrieves risk clauses in milliseconds, validates citations, and outputs interactive reviews without cloud leakage.

Problem

Freelancers and small agencies need to review contracts for risky clauses but cannot afford legal review and cannot upload confidential documents to commercial cloud LLMs.

Outcome

VaultDocs AI provides a privacy‑guaranteed, zero‑cloud contract risk screening workflow that runs locally, reduces cost to $0 per query, and delivers risk scores and audit reports without exposing confidential data.

Workflow

• Parse PDF contracts into page‑bound chunks using PyMuPDF.
• Rank clauses with an in‑memory BM25 + legal keyword hybrid ranker and FAISS vector index to retrieve risk‑prone sections in under 5 ms.
• Screen retrieved sections locally with open‑weight LLMs (llama3.2:1b, gemma:2b, qwen2.5:3b) via Ollama.
• Validate model citations against extracted PDF text to eliminate hallucinations.
• Render a professional 3‑column legal review workstation with customizable risk scoring (0–100) and downloadable Markdown audit reports.

Evidence

VaultDocs AI is a local-first, privacy-guaranteed contract risk screening assistant tailored for freelancers, contractors, and boutique agency owners.

Source

No contract text, vector embeddings, or prompts ever leave local memory.

Source

Parses PDF contracts into page-boundary chunks.

Source

Uses a fast in-memory hybrid ranker (BM25 + legal keyword boost) and dense FAISS vector index to retrieve risk-prone sections in under 5 milliseconds.

Source

Performs local open-weight LLM screening using llama3.2:1b, gemma:2b, or qwen2.5:3b via Ollama.

Source

Validates model citations directly against extracted PDF text to eliminate hallucinations.

Source

Renders a professional, interactive 3-column legal review workstation with customizable risk scoring (0–100) and downloadable Markdown audit reports.

Source