LexGuard — AI Contract Risk & Deadline Manager

• 

  LexGuard: AI contract risk analysis built in Rust. Argon2id hashing, SHA-256 sessions, per-IP rate limiting.

• 

  Registration enforces 12+ char passwords, rejects duplicate emails, hashes with Argon2id before first write.

• 

  Fresh account, zero data. Clean dashboard state — proves onboarding works without seeded or fake data.

• 

  Live dashboard: 3 contracts, 2 pending reminders, 3 AI analyses, aggregated from Postgres in real time.

• 

  Contract list never leaks raw_text or other users' data. Ownership is enforced in SQL, not just the UI.

• 

  Reminders follow a 7/3/1/0-day schedule per obligation. Regeneration is idempotent, no duplicate spam.

• 

  Discord channel connected. Webhook secrets are encrypted at rest with ChaCha20-Poly1305, never logged.

• 

  Profile is read-only by design. No endpoint ever exposes another user's ID, email, or contract data.

• 

  Sessions are opaque tokens hashed with SHA-256 server-side. Logout revokes the session, not just the cookie.

• 

  Mutual Service Agreement: 0/100 Low. The AI reports "no material risks found," not a false "this is safe."

• 

  Consulting Agreement: 70/100 High. Three named risks, each traced back to the exact clause that caused it.

• 

  Every risk quotes real contract text, fuzzy-matched against the source to catch AI hallucination before it ships.

• 

  Employment Agreement: 90/100 Critical, 11 risks. Same Groq model, same prompt, genuinely different verdict.

• 

  Top risk: a 50-year global non-compete, scored 90/100 Critical, with the exact clause quoted as evidence.

Inspiration

Contracts hide critical deadlines and one-sided clauses in dense legal language. Missing a termination notice or an unfavorable liability clause can cost real money — but most people and small businesses can't afford a lawyer to review every agreement. We wanted an AI tool that flags risky clauses with evidence, not just a vague "this looks risky."

What it does

LexGuard lets you upload a contract (paste text, PDF, or a photo via OCR) and automatically:

• Detects and scores risks (e.g. asymmetric liability caps, one-sided termination notice, unilateral amendment rights)
• Extracts obligations with responsible party and due dates
• Backs every finding with an exact quote from the contract as evidence
• Schedules automated reminders (7/3/1/0 days before deadlines) delivered via Discord/webhooks

How we built it

• Backend: Rust + Axum, PostgreSQL with SQLx migrations
• Frontend: Leptos compiled to WebAssembly
• AI: OpenAI-compatible chat completion (served via Groq) for contract analysis
• Security: Argon2id password hashing, opaque session tokens, ChaCha20-Poly1305 encrypted webhook credentials, SSRF protection on outbound webhooks
• A background worker handles reminder delivery with retries and crash recovery
• Fully containerized with Docker, 333+ automated tests

AI disclosure

Two separate uses of AI here: the app itself uses AI at runtime (an OpenAI-compatible provider — Groq/OpenAI/Gemini — for contract risk analysis and image OCR, described above). Separately, as the developer, I used Gemini and DeepSeek for coding assistance and Claude for code review, security hardening, and architectural feedback while building this solo in 8 days. All architecture, integration, testing, and final review were mine. Full day-by-day disclosure: DEVELOPMENT_PROCESS.md.

Challenges we ran into

LLM outputs are non-deterministic — the same contract could get slightly different risk scores across runs. We built a fuzzy evidence-verification system that checks AI-cited evidence against the original contract text (exact match or 70% token overlap) to catch hallucinated citations before showing them to the user.

What we learned

Building trust into AI outputs matters more than raw accuracy — surfacing verifiable evidence for every claim, and being transparent about known limitations, is what makes an AI legal tool usable.

What's next

Jurisdiction-aware analysis, contract-type-specific risk models, and support for more AI providers.

Built With

• api
• argon2id
• axum
• chacha20-poly1305
• discord
• docker
• groq
• leptos
• openai
• postgresql
• rust
• sqlx
• webassembly
• webhook