- Challenge
- A hard-coded schema chatbot broke on 100k+ CDCR records and compound queries. Experts needed a workspace, not a chat toy.
- Approach
- We built a Neural Query Engine (Text-to-SQL/Pandas), Scenario Builder, Bias Analysis (RJA), Justice Reform Q&A (RAG), Case Explorer, dashboard, embedded Power BI, and governance/data dictionary.
- Result
- 100,000+ records (91,840 cases in 2025 dataset experience). Analyses that took days now run in seconds. Used by Stanford Three Strikes Project and Dream.Org. Methodology cited in 2025 research. AI scoped to NL-to-query and summarization (no legal judgments).
Redo.io
Criminal justice data analytics
Legal / justice reform analytics (public defenders, researchers, journalists)
Related on Beyond AI Cloud
Related work
Agentive
Agentic GTM / revenue orchestration
Early-stage B2B GTM / revenue SaaS (paid pilots)
ICP, outreach, pipeline, and account engagement lived in disconnected tools. Handoffs were manual and deals went quiet.
LegalAI
Private on-device legal document intelligence
Legal professionals / solicitors; confidential document review
A dual trust problem: fully private (no cloud or telemetry) and zero invented or missed clauses across many document types. Cloud AI is unsuitable for confidential legal review.
Sannos / Evolve
RAG compliance automation (CSRD, DORA, ESG)
GRC / compliance / ESG
The intern pipeline was not real RAG (~40 min/doc), sequential, and untrustworthy. Everything had to stay inside locked-down Azure.