Inspiration
Wrongful convictions affect thousands of individuals, yet reviewing parole hearing transcripts for bias and innocence claims is tedious and time-consuming. We wanted to leverage AI to help legal advocates quickly identify cases with the strongest potential for exoneration.
What it does
JusticeLens is an AI-powered legal case analysis platform that:
- Analyzes parole hearing transcripts to extract key case information
- Detects commissioner bias with severity scoring and quoted evidence
- Calculates innocence scores based on claims, evidence quality, and wrongful conviction indicators
- Generates structured legal memos with case details, people mentioned, and potential theories
- Enables team collaboration with case assignment, comments, and a Kanban pipeline
How we built it
- Frontend: React, TypeScript, Tailwind CSS, shadcn/ui
- Backend: Lovable Cloud with PostgreSQL database and Edge Functions
- AI Analysis: Gemini/GPT models for transcript parsing, bias detection, and memo generation
- Real-time: Live updates for team collaboration features
Challenges we ran into
- Accurately detecting nuanced commissioner bias from transcript language
- Balancing innocence score weighting across different evidence types
- Building a flexible pipeline that adapts to different legal workflows
Accomplishments that we're proud of
- Comprehensive bias detection with severity levels and direct quotes
- Multi-module analysis system (legal memo, bias, people extraction)
- Team invite system with shareable links for collaboration
What we learned
- How to structure AI prompts for reliable legal document analysis
- Importance of RLS policies for multi-team data security
- Building accessible UI for complex legal workflows
What's next for JusticeLens
- Email notifications for case updates and mentions
- Integration with legal case management systems (Lawmatics)
- Expanded AI analysis for additional document types
- Commissioner tracking across multiple cases for pattern detection
Built With
- cloud
- gemini
- lovable
- supabase
- toolhouse
Log in or sign up for Devpost to join the conversation.