block
columns 1
product["<b>Your AI product</b> · agents, copilots, document AI"]
runs<["runs on"]>(down)
l1["<b>01 · Serving & routing</b><br/>multi-GPU vLLM on Kubernetes · LiteLLM gateway<br/>one-switch provider portability (local ↔ cloud)"]
l2["<b>02 · Retrieval you can trust</b><br/>pgvector RAG · corrective fallback, fail-closed<br/>embedding-parity gates against silent drift"]
l3["<b>03 · Evals, observability, spend</b><br/>CI-gated evals (pure metric + LLM judge)<br/>retrieval alarms · cost ceiling with kill-switch"]
l4["<b>04 · Agents in production</b><br/>MCP tooling · human-approval gates · guardrails"]
l5["<b>05 · Multi-tenant cloud underneath</b><br/>tenant isolation · CDK / Terraform · OIDC deploys"]
classDef mine fill:#0969da,stroke:#0550ae,color:#ffffff
classDef base fill:#57606a,stroke:#424a53,color:#ffffff
classDef yours fill:none,stroke:#8c959f,stroke-dasharray:4 3
class l1,l2,l3,l4 mine
class l5 base
class product yours
| Years | Role | Proof |
|---|---|---|
| 2026 – now | Principal · multi-tenant AI SaaS | tenant isolation, AI spend ceilings, Document AI on AWS |
| 2024 – 2026 | Principal · GenAI infrastructure | multi-GPU vLLM on Kubernetes: ~25% lower cost, ~40% lower latency · pipelines serving 80+ app vendors |
| 2017 – 2024 | Senior · national digital-health platform | 100k+ users, 99.5% uptime · OAuth 2.0 SSO for 25,000+ users, SAML across 5+ partner orgs · zero-downtime AWS migration · deploys 4 h → 20 min |
| 2010 – 2017 | Engineer → Lead · ERP, school MIS, hospitality | led microservices architecture and sprint delivery · SOAP → REST modernization · engineering standards that cut production incidents 65% |
| Repo | What it shows |
|---|---|
| kattakath/skills | Claude Code plugins and skills — agent tooling, guardrails, MCP |
| kattakath/ircc-whatsapp-bot | Corrective RAG (LangGraph + pgvector) answering only from official sources |
| kattakath/nix-config | Fleet infrastructure as code — CI gates, terranix, secrets, observability |
| kattakath/.github | Shared CI landing system — one required gate, merge queue, reusable workflows |
Stack: Python · TypeScript · FastAPI · Next.js · AWS · GCP · Kubernetes · vLLM · LiteLLM · pgvector · Terraform · Nix



