Case Studies
Short case studies from my AI architecture work: the problem, the boundary, the tradeoff, and the operating signal I watch.
systems
5
Architecture stories, not mockups
production
3
Built from real delivery patterns
runtime graph
29/28
Runtime nodes and edges
Case Study 01
Python FastAPI gateway
I built a local OpenAI-compatible gateway so my site, agents, and scripts could call one stable /v1 contract. Ollama, PostgreSQL, Redis, and Qdrant stay private on the Mac. The hard parts were auth, backpressure, model warmup, and zero public inbound ports.
Ingress
Cloudflare
Gateway limit
10 chats
Model runtime
Ollama 0.30.6
Memory profile
q8 KV
Case Study 02
BFSI workloads
I designed an agentic RAG pattern for long financial filings. The system needed layout-aware parsing, hybrid retrieval, clear state transitions, and a grounding check before any answer could ship.
Retrieval
pgvector
Orchestration
LangGraph
Runtime
GKE/Istio
Reliability
Eval gate
Case Study 03
Inference scaling
I worked on inference platform patterns where static GPU allocation slowed teams down. Production serving needed quota, priority, and predictable capacity.
Scheduler
Run:AI
Serving
vLLM
Cluster
OpenShift
Signals
OTel
Case Study 04
Upskilling teams
I designed patterns for teams adopting MCP-style tools. The goal was to let agents call databases and APIs without exposing raw credentials or ownership boundaries.
Protocol
MCP
Security
Sandbox
Governance
Schemas
Adoption
Playbooks
Case Study 05
Static Git Graph RAG
I wanted fast site retrieval without a managed vector database. The system needed low token use, fast lookup, and a Git-friendly build path.
Public Surface
Next.js /api/retrieve
Trie Search
MiniSearch
Retrieval Speed
< 0.5ms
Token Savings
80-90%