AI discovery platform for criminal defense

Defense attorneys receive discovery as thousands of unsorted files: police reports, bodycam and interview recordings, emails, bank and subpoena returns. A single federal case can run past 80,000 pages. Reading it by hand is slow and expensive, and a missed name or date can decide the outcome. CrossLogic's founders, former detectives and forensic examiners, needed a platform that reads every file, builds the timeline and the cast of people, and answers questions with a citation back to the source.

1. 1Ingests from browser upload or directly from Box, Dropbox and Google Drive, with parallel import lanes and duplicate detection.
2. 2Parses and OCRs documents and emails, and transcribes bodycam, interview and audio recordings onto the same case record.
3. 3Extracts people, organizations, locations and dated events, then merges them into one case-wide timeline and entity graph.
4. 4Answers questions through hybrid retrieval (pgvector plus a keyword index), with every answer citing the document it came from.
5. 5Meters AI cost per case and per stage, with per-case budgets and model routing set from an admin panel.

• An 83,000-page federal case fully extracted at about $0.003 per page in AI cost.
• Single Box imports of 110,000+ files, with duplicates caught before they cost anything.
• AI spend metered per case and stage, reconciling with the provider's bill to the cent.

• FastAPI
• Next.js
• PostgreSQL + pgvector
• Celery + Redis
• OpenRouter (multi-model)
• Docker

• legal tech
• rag
• ai extraction
• multi-tenant saas

• Back-office automation and custom AI agents — Custom back-office automation and AI agents for operations and finance teams: invoicing, reports, ERP syncs and chat over your tools. One senior engineer.