Case study · Spire Grounds · Jan 2026 – present
Spire Training Group — AI platform
FastAPI + React AI platform for training — RAG and live session direction.
- Role
- AI Engineer & Data Architect
- Building
- AI Training Platform
- Year
- Jan 2026 – present
Problem
Training needs an inspectable AI platform that can answer domain lookups quickly, support training-data workflows, and direct live Unity sessions over the network.
Daniel's role
Built a FastAPI + React AI platform with 30+ APIs; RAG with pgvector; team-builder models; a streaming chat endpoint with parallel RAG retrieval that cut lookup latency from about 9–28s to about 6.2s; migrated lookups from GPT-4o-mini to Claude; built an SFT/DPO training-data pipeline; and an adaptive scenario director: a WebSocket service that runs Unity sessions for training.
System diagram
Stack
- FastAPI
- React
- pgvector
- Claude / OpenAI APIs
- WebSockets
- Unity session direction
Outcome
Shipped a multi-API AI platform with a streaming chat endpoint and parallel RAG that brought lookups to about 6.2s, a Claude migration, team-builder models, an SFT/DPO data pipeline, and an adaptive scenario director for Unity sessions.
Describe architecture and impact clearly for training deployments without exposing restricted customer detail.
Field notes (measurements)
APIs
30+
Lookups
~6.2s