~/About~/Systems~/Foundry~/Blueprint~/Journal~/Projects
Book a Call

Multi-Agent RAG Platform

The complete system record. Choose the proof lens that fits your role.

View on GitHub

Choose The Proof Lens

One system record, three reader paths.

Feature-frozen3-act recordGitHub available
Business proofFounder / operator

The AI Backend That Every Document-Heavy Team Rebuilds From Scratch

Document-heavy teams get grounded AI search, hybrid retrieval, guardrails, and cost budgets without rebuilding RAG every 3-4 months.

System FlowData ModelArchitecture LayersDecision Log
Open business proof6 min read
Architecture proofCTO / architect

Architectural Brief: Multi-Agent RAG Platform

Neo4j made optional because it won't fit in a 1GB VPS. Hybrid retrieval weighs vector, keyword, and graph at 0.7, 0.2, 0.1 with graceful degradation.

System TopologyInfrastructure DecisionsConstraintsDecision Log
Open architecture proof6 min read
Technical proofSenior engineer

I Built a Knowledge Graph Into the Retrieval Pipeline and Then Dropped It in Production

The knowledge graph added two hundredths of a point to retrieval relevance for 512MB of extra memory. In production, it runs behind a flag.

Code SampleFailureConstraintSurprise
Open technical proof7 min read

Related Systems

More proof in the same problem space.

View systems map

Contract Lifecycle Engine

AI extraction with confirmation

Strongest example of keeping AI-generated obligations in Pending until a human confirms them.

Centralized Property Intelligence Hub

Searchable operational knowledge layer

Shows how replicated property data became a queryable intelligence layer without calling four services per question.

NBA Scenario Engine

Scenario modeling from incomplete evidence

Useful proof for modeling around survivor bias and missing data instead of pretending the dataset is clean.