AI Developer Lead @ Minfy · Founder, TechNaom AI Labs (independent GenAI R&D)
GenAI Solution Architecture · Agentic AI · RAG · LLMOps · SRE
10+ years in software, now leading AI development at Minfy. I architect and ship enterprise Generative AI — RAG systems, agentic pipelines, LLMOps and deterministic automation that save real engineering time, from architecture through deployment.
I run TechNaom AI Labs, my independent GenAI R&D lab, where I build in the open: agents on open-weight models, MCP servers, eval harnesses, and autonomous workflows that run in production on a $10/month budget.
| AI agents in my digital workforce | engineer-hours saved / year | reviews automated / year (4×) | prompts in enterprise library | awards incl. 1 Platinum |
| Role | Where | Focus |
|---|---|---|
| AI Developer Lead | Minfy — AWS / cloud-native AI systems integrator | Building production GenAI and agentic systems on AWS (Bedrock, AgentCore, LangGraph) |
| GenAI Solution Architect | ValueMomentum — P&C insurance-focused IT services | GenAI solution architecture for property & casualty insurance use cases |
| AVP, Senior Analyst | Bank of America | Enterprise GenAI adoption, compliance automation, SRE — 28 awards incl. Platinum |
| Founder | TechNaom AI Labs | Independent GenAI R&D: agents, MCP, evals, open research and free courses |
- 🧪 Open research — making Gemma 4 on a single accelerator a reliable SWE agent (gemma4-agentic-swe, Google DeepMind × Kaggle paper track)
- 🪢 Open source — contributing to Langfuse LLM observability (contribution log)
- 🤖 Autonomous AI in production — a self-running daily content pipeline with hard cost caps and full tracing (field guide)
- 📚 Teaching — the free For Everyone series: RAG, Agents, MCP, Evals, AI Security and more
| System | What it does | Impact |
|---|---|---|
| Compliance automation framework | Structured prompt-engineering system covering a 4-phase review lifecycle and 15 audit controls | 40 → 10 min per review · 28,600+ reviews/yr · ~14,300 hrs saved |
| Deterministic rule engine | Re-architected into a schema-gated JSON rule catalog — 130 rules, zero LLM calls at runtime, zero data egress | Full audit traceability · ~95% faster turnaround |
| Production RAG knowledge base | Recurring team Q&A turned into semantic search (LangChain · ChromaDB · Streamlit) | Serves 100–150 practitioners on demand |
| Enterprise Copilot rollout | GitHub, Teams and M365 Copilot adoption; founded the prompt-engineering team | Mentored 8–10 teams → 80+ of 200+ automations |
|
Open research: turning Gemma 4 on one accelerator into a reliable software-engineering agent. Google ADK, code-graph tools, SWE-bench-style scoring, every number published — including what didn't work.
|
How an autonomous AI workflow runs daily in production: serverless, $10/month hard cap enforced in code, ~750 tests, kill switch, idempotent publishing, full Langfuse tracing.
|
|
Multi-agent design (planner → scouts → curator → writer → critic) on self-hosted open-weight models, built to measure whether agents beat a fixed workflow per dollar.
|
Production-oriented advanced RAG: hybrid retrieval, reciprocal rank fusion and LLM re-ranking. Plus rag-projects — end-to-end RAG with evaluation, built from scratch.
|
My lab runs 36 AI agents built in under two months, on track to 200. They research, write, review, evaluate and publish — every post on my LinkedIn is generated and shipped by them, with quality gates, cost caps and full tracing.
Private builds (code private, happy to walk through the architecture):
| Project | What it is |
|---|---|
| 🏗️ ai-agent-platform | Production multi-agent + multi-RAG platform with first-class evaluation and observability |
| 📄 doc-intel-mcp | Document-intelligence MCP server on AWS: catalog, section tree and hybrid index for agents |
| 📰 linkedin-branding-engine | Daily AI-news → LinkedIn pipeline: Tavily top-10, cost-tiered LLM writer, quality gate, publisher |
| 🕸️ agentic-content-engine | LangGraph agents that write grounded posts — the build behind the public design |
| 📣 content-promotion-engine | Multi-channel distribution (LinkedIn, X, Reddit, Dev.to, Discord) with UTM tracking and human-gated engagement |
| 📊 linkedin-insights-engine | Weekly agentic flow analysing what makes AI and leadership posts perform |
| 🔍 ace-review-bench | R&D bench comparing AI code reviewers (Greptile, Cubic, CodeRabbit) on real bugs |
| 🛡️ HackerRank Orchestrate (Oct 2026) | Deletion-audit agent that finds "ghost data" that still exists after deletion |
| 🐍 Python · Live → | 🤖 GenAI (7-week program) | 🔎 RAG |
| 🦾 Agentic AI | 🔌 MCP | 🧭 Context Engineering |
| 📏 LLM Evaluation | 🏭 AI Engineering | 🛡️ AI Security |
| 💻 AI Coding Agents | ♾️ DevOps | 🌱 git-actually — real Git + CI/CD practice |
Core strengths: GenAI solution architecture · agentic systems (LangGraph, AWS Bedrock, AgentCore) & MCP · RAG & vector search · AI evals, AI security & AI governance · LLMOps & observability · SRE & incident response (RCA, SLO/SLI) · Python engineering · AWS & CI/CD
28 awards in 5 years at Bank of America — including 1 Platinum, the bank's highest delivery recognition, for GenAI leadership and enterprise AI adoption, plus a Gold for enablement work.
Open to conversations on GenAI architecture, agentic systems, LLMOps and enterprise AI adoption. Reach me at technaom@gmail.com, on LinkedIn, or at manoharpapasani.pages.dev.
