Evidently

• 

  Evidently

Inspiration

Evidently started because AI could help me investigate a sprawling evidentiary record, but its context window could not hold the case. I needed a backend to preserve the complexity so AI could focus on the task in front of it.

My first attempt used Google Antigravity to move facts into a database. It was great to watch and completely unusable. That became Evidently.

What it does

Evidently is an investigation, case, and evidence management system for AI-assisted work.

It preserves original electronic evidence beyond reach, in the vault, while investigators and AI work across annotations, facts, entities, events, relationships, chronology, contradictions, and inferences.

AI cannot alter underlying evidence. Inferences must trace to supporting evidence, absent user override.

From an investigator building a timeline to a prosecutor drafting a motion, specialized AI can interface with and operate across Evidently.

How I built it

I built Evidently so persistent case complexity lives in the backend.

Original evidence remains separate from analysis. AI receives the relevant case context for a particular task while Evidently preserves provenance, relationships, and accumulated case knowledge.

Challenges I ran into

Context windows limit how much of a complex case AI can reason over at once.

Poorly structured intake also makes extracted information useless. High-stakes work adds another constraint: hallucination control is built in with the auditability of inference mapping.

Accomplishments I'm proud of

I made AI a first-class participant without giving it control over evidence.

Evidently preserves case knowledge across sessions, keeps inferences tied to evidence, and allows different AI systems to work through the same evolving record.

What I learned

AI works better when the backend carries the structure. Persistent case knowledge lets models focus on where needed while preserving years of accumulated investigative work.

What's next for Evidently

I am seeking funding to move Evidently from a working system into production: harden security, expand agent integrations, scale evidence processing, and put it in the hands of investigators and attorneys working real cases.

Google Cloud and Gemini give me an immediate path to test Evidently at scale while extending its architecture.

Built With

• amazon-web-services
• cloud-run
• docker
• fastapi
• gemini
• google-adk
• google-cloud
• mcp
• python
• rag
• rest-api
• sqlite
• vector-search
• vertex-ai