ClauseGuard AI
In ProgressAI-powered contract risk review for freelancers and small agencies — isolating dangerous clauses in seconds and translating legalese into actionable insight.
ClauseGuard AI is an intelligent contract analysis platform engineered specifically to protect independent workers from predatory, vague, or financially damaging contract clauses. By combining natural language processing with automated risk scoring, the platform isolates dangerous terms in seconds and translates legalese into actionable insights.
The global gig economy generates roughly $550–582 billion in annual revenue. The U.S. freelance workforce grew 90% between 2020 and 2024, with projections of 86.5 million U.S. freelancers by 2027. Despite this growth, independent workers remain acutely vulnerable to contractual exploitation.
ClauseGuard AI democratizes contract review by automating the extraction and risk-scoring of critical clauses — Payment Terms, Scope of Work, Intellectual Property, Non-Competes, and more.
A modular, event-driven architecture built to balance performance, scalability, and strict data privacy.
Minimalist, value-first interface built with Tailwind CSS and shadcn/ui
PDF extraction service, with a Zod validation layer in front of it
Relational tables for contracts & clauses, plus pgvector for rubric embeddings
Async webhooks → aggregation → notification, without blocking the main backend thread
RAG pipeline producing structured JSON output
Given the sensitive nature of legal documents, ClauseGuard AI follows fundamental cybersecurity practices aligned with the CIA Triad — Confidentiality, Integrity, Availability.
- Strict MIME-type validation on all uploads
- Prompt injection sanitization before documents reach the model
- Secure storage in Supabase with aggressive data-retention lifecycle rules
ClauseGuard AI doesn't just summarize documents — it runs a classification pipeline using Retrieval-Augmented Generation. Standard legal rubrics are embedded into a pgvector database, so the Gemini model grounds its analysis in established legal thresholds before assigning a risk profile.
This ensures highly critical clauses — like a perpetual IP transfer or an aggressive non-compete — dramatically elevate the risk score, prompting immediate review via highlighted Risk Cards.
0–1
2
3
4
5
- Enforcing AI structured output — free-form LLM text is brittle for legal analysis, so strict JSON output via Gemini enables deterministic mapping directly into PostgreSQL.
- Asynchronous orchestration — parsing and classifying 20-page contracts synchronously risks API timeouts, so the pipeline is offloaded to an event-driven n8n workflow.
- Preventing hallucinations via vector search — pgvector retrieves trusted "good" and "bad" clause examples before prompting Gemini, reducing false positives.
Beyond the initial sprints, the long-term roadmap includes OCR for scanned documents, fine-tuned legal models, a Word plugin integration, and multi-jurisdictional rubrics to support freelancers operating globally.