Investigator
Autonomous AI intelligence platform that conducts multi-hour investigations, discovers hidden relationships, and generates comprehensive intelligence reports, all without human intervention after the initial query.
Research that eats weeks
Investigative journalists spend weeks tracing corporate ownership structures.
Due diligence teams manually comb through thousands of documents looking for red flags.
Legal professionals burn hundreds of billable hours on discovery.
What if an AI could do the research autonomously while you focus on the insights?
Inspiration
The launch of Gemini 3, with its 1M token context, Thought Signatures, and Live API, made this possible for the first time.
I was not interested in building another chatbot. The goal was an autonomous agent that thinks like an investigator: working for hours without supervision, backtracking when evidence contradicts a hypothesis, and showing its reasoning every step of the way.
Built to run alone
An autonomous engine that discovers, analyses, and visualises complex information networks in real time.
Investigation Board
Real-time knowledge graph visualisation with auto-layout algorithms. Interactive entities and relationships that build themselves as the AI discovers connections.
Entity Discovery
Autonomous identification of people, companies, locations, and events. Each entity is extracted with confidence scoring and relationship mapping.
Evidence Tracking
Source credibility assessment and evidence linking. Every claim is backed by traceable evidence with a full provenance chain.
Thought Chain Transparency
See the reasoning step by step. Watch hypotheses form, get tested, and evolve as new evidence emerges, instead of trusting a black box.
Visual evidence
Command Center
Central intelligence hub with real-time investigation monitoring and analytics.
Active Operations
Multi-case management with priority indicators.
Case File Analysis
Entry Point
Secure Login
Registration
Tech stack
Production architecture combining async task orchestration, real-time WebSockets, graph algorithms, and reliable AI integration.
Backend
- Django REST Framework
- Celery
- Redis
- PostgreSQL
- Django Channels
- NetworkX
Frontend
- Next.js 14
- TypeScript
- React Flow
- Tailwind CSS
- shadcn/ui
AI / ML
- Google Gemini API
- Thought Signatures
- Multi-step Reasoning
- Entity Extraction
Infrastructure
- WebSockets
- Real-time Updates
- Graph Algorithms
- Async Task Queue
Challenges overcome
Real distributed-systems problems, and how I solved them.
Celery task not triggering
Investigations stayed stuck on pending. The Celery task call was commented out as a TODO. The simplest bugs are the hardest to spot.
Nodes stacking at (0, 0)
Integrated a NetworkX spring layout (k=2, iterations=50, scale=1000) so the graph lays itself out cleanly instead of piling on the origin.
Gemini API returning empty results
The API key was an empty string. Added validation, better error handling, and loud logging for the integrations that matter most.
WebSocket authentication
Built custom middleware to pull the JWT from query parameters, since WebSockets cannot send Authorization headers.
Race conditions in the graph
Used Django transactions for atomic saves, broadcast entity creation before relationships, and queued updates on the frontend.
What's next
Immediate roadmap
- Voice integration with the Gemini Live API
- Multi-modal evidence analysis
- Collaborative investigations
- Advanced network analytics
Future vision
- Investigation templates
- External data connectors
- Pattern detection with ML
- Enterprise features and SSO
Interested in collaboration?
This project is autonomous AI agents, distributed systems, and real-time visualisation in one build. Let us talk about what that can do for you.