-6366F1?style=for-the-badge&logo=shield&logoColor=white){kind=icon}
| 🧩 Vectors & Chunks | 💰 Monthly Cost | ⚡ Retrieval | 🛡️ MCP Trust Score | 🌟 Open Source Reach | 🤖 Live Platforms |
|---|---|---|---|---|---|
| 32,000+ | ₹0 / month | 183 ms | 89/100 (M8ven) | 100+ Stars · 29+ Forks | 4 Live Platforms |
Four production AI platforms. Zero cloud spend. Real users. Real queries.
FastMCP tools, Agentic RAG, and fine-tuned LLMs. Deployed, documented, and battle-tested.
Official Global Spotlight · M8ven Verified MCP Publisher · 100+ GitHub Stars · Published Technical Dossiers
| # | Technical Case Study / Publication | Platform | Focus Area | Impact & Validation |
|---|---|---|---|---|
| 1 | High-Reliability Agentic RAG on 512MB RAM | UptimeRobot Official Spotlight | Zero-Cost Infrastructure & Container Recovery | Official Global Feature · 99.998% SLA |
| 2 | 11-Node Production Agentic RAG Playbook | Medium Publication | LlamaParse VLM, Matryoshka MRL & FastMCP | Published Engineering Breakdown |
| 3 | Autonomous Financial Parser (Show HN) | Hacker News / GitHub | Production Multi-Agent Retrieval | 100+ Stars · 29+ Forks |
| 4 | Evidence-Grounded Career Workspace | M8ven MCP Trust Index | JSON-RPC Tool Calling & Schema Security | 89/100 Security & Schema Audit |
| 5 | Production RAG Constitution & Field Guide | Engineering Docs | 4-Layer OOM Shield, Parent-Child Sync | Full Technical Dossier |
Selected & Interviewed by UptimeRobot for architecting high-reliability Agentic RAG systems (99.98% uptime) and solving container sleep + database pauses on zero-cost 512MB RAM infrastructure.
🔗 Read the full published case study: UptimeRobot Blog — Community Spotlight: Ambuj Kumar Tripathi →
✍️ Check out the detailed engineering breakdown: Medium — Building an 11-Node Production RAG System →
M8ven MCP Trust Index: Audited and verified publisher with 89/100 security & schema trust score on the public MCP registry. Open-Source Community Reach: 100+ GitHub Stars and 29+ forks driven by Show HN community launch for the Agentic RAG platform.
Next.js 16FastMCPGemini AILangfuseLangSmithMongoDBGoogle OAuthRechartsTailwindCSS
🔍 Core Engine → Evidence-Grounded Resume Claim Validation & Interview Defense Simulator
⚡ FastMCP Tools → JSON-RPC FastMCP Server with Live Jina Reader & Tavily Company Web Research
📊 Scoring Radar → 5-Axis Competency Fit (Technical, Domain, Seniority, Tools, Culture)
🛡️ MCP Audit Gate → Verified Publisher on M8ven Trust Index (89/100 Security & Schema Audit Score)
🧠 LLMOps Tracking → Dual Langfuse & LangSmith Tracing for Token Costs, Latency & Failure Diagnostics
🔒 Authentication → NextAuth Google OAuth 2.0 with MongoDB 30-Day TTL Session Persistence
💰 Infrastructure → Live Production on Vercel · Server-Sent Events (SSE) · Zero Cold-Start
LangGraphPineconeFastAPIJina v3 MRLLlamaParse VLMCohere RerankerGemini 3.5 Flash LiteMongoDBRedis/UpstashLangfusePresidioSupabasepybreaker
📚 Knowledge Base → Budget 2024-25, Finance Bill, Tax Laws, RBI Guidelines, Constitution
🧩 Vector Scale → 14,662 vectors · 100+ GitHub Stars & 29+ Forks · Jina v3 MRL (1024→256d)
📄 Document Parse → LlamaParse VLM — vision-language model for complex tables & layouts
🧠 Orchestration → 11-Node LangGraph StateGraph · Classifier (6-path) · Parallel Retrieval
❓ Tools & HITL → Web Search + Stock Tool (Gemini) · HITL clarification & permission
🛡️ LLM-as-Judge → Hallucination Guard — separate LLM call verifies grounding before response
🔒 Security → 7-Layer Upload Pipeline + PII Shield (regex) + Circuit Breakers (pybreaker)
🔄 Sync Engine → SHA-256 idempotent upserts · Zero duplicate vectors · Zero ghost chunks
💰 Infrastructure → ₹0/month on Render free tier · WhatsApp Bot · Daily AI Newsletter
LangGraphQdrantFastAPIJina AIQwen 3 235BMongoDBRedis/UpstashLangfusePresidioSupabase
📚 Knowledge Base → 6 Indian Legal Acts (IPC · BNS · BNSS · BSA · PCSO · IT Act)
🧩 Chunking → 10,833 child vectors · Parent-Child (400-char search / 2000-char LLM context)
⚡ Performance → 183ms retrieval · Sub-300ms end-to-end · SSE streaming responses
🛡️ Security → PII masking (Aadhaar / PAN / Mobile) before embedding · GDPR 30-day TTL
🔄 Sync Engine → SHA-256 idempotent upserts — zero duplicate vectors, zero ghost chunks
🧠 Orchestration → LangGraph 6-node state machine · 3 runtime paths (RAG / Greeting / Abusive)
🎯 Quality Gate → Confidence threshold at 40% cosine — zero hallucinated legal citations
💰 Infrastructure → ₹0/month on Render free tier · 99%+ uptime
LangGraphChromaDBFastAPIOpenRouterSlowAPIReactVercel
💬 Conversations → 10,000+ real user conversations processed
🔢 Tokens → 525,000+ tokens handled in production
🔒 Rate Limiting → SlowAPI — 5 req/min per IP (burst traffic protection)
⚡ Resilience → Circuit Breaker: fail_max=10, reset_timeout=120s (cascade failure prevention)
🐛 Production Fix → ChromaDB 0.6.x telemetry deadlock → pinned chromadb==0.4.24
💰 Infrastructure → ₹0/month on Vercel + Render free tier
Orchestration & Backend
Vector DB & Storage
LLMs & Embeddings
LLMOps & Security
Frontend & Deployment
Hand-coded SVG · Animated flows · Production Data
📐 Architecture Breakdown — Click to expand ▾
| Layer | Components |
|---|---|
| Ingestion | PDF Loader → LlamaParse VLM → 4-Layer OOM Shield → Parent-Child Chunker → Jina AI Embed → SHA-256 Sync → Pinecone/Qdrant Upsert |
| Query Entry | React → FastAPI → Google OAuth → Presidio PII Mask → Redis Cache Check |
| LangGraph | classify_node → 4 runtime paths (RAG / Greeting / Vague / Abusive) |
| RAG Path | retrieve_node → generate_node → Hallucination Guard → post_process_node |
| HITL | CrossQuestioner node → 2-round clarification for vague queries |
| Persistence | Pinecone · Qdrant · MongoDB · Redis/Upstash · Langfuse · Supabase · Circuit Breaker |
Complete technical documentation of all production systems — architecture decisions, failure logs, chunking strategies, OOM prevention, LangGraph state machine deep-dives.
Topics covered: Document Loaders · LlamaParse VLM · Parent-Child Chunking · SHA-256 Sync Engine · LangGraph StateGraph · HITL CrossQuestioner · Hallucination Guard · OOM Prevention · Adaptive Retrieval · Deployment Failures & Fixes
Sep 2025 – Oct 2025 | Remote, India (Contract — Concluded due to mandatory office relocation requirement)
✓ 40% reduction in prompt iteration cycles for production LLM applications
✓ Built 25+ reusable prompt libraries across image and text generation models
✓ Systematic model validation, adversarial testing & QA pipelines
Flux SDXL Prompt Engineering Model Validation Adversarial Testing
Jan 2022 – Aug 2024 | Gurugram, India
✓ Led FTTP network planning (GIS, PipeR, Amanda toolchain)
✓ 98% QA compliance — "Top Performer" award
✓ Automation reduced manual effort by 70%
| Company | Role | Period | Impact |
|---|---|---|---|
| Tata Communications | O&M Engineer | Nov 2021 – Jan 2022 | 10% downtime reduction |
| TCS | System Admin | Oct 2013 – Nov 2014 | 99% uptime for revenue systems |
| Annu Infra | Fiber Engineer | Dec 2017 – Jun 2018 | Defense-grade fiber projects |
| Provider | Areas |
|---|---|
| NVIDIA | RAG Agents with LlamaIndex · Building RAG Agents · Jetson Nano AI |
| Google Cloud | Gemini · Vertex AI · TensorFlow · Responsible AI |
| IBM | AI Fundamentals · Deep Learning · Generative AI |
| Microsoft Azure | Responsible AI · Azure AI Fundamentals |
| Forage / Industry | BCG · AWS · Deloitte · Tata simulations |
| Degree | Institution | Year |
|---|---|---|
| PG Diploma — Power Transmission & Distribution | NPTI, Delhi | 2015–16 |
| B.Tech — Electrical & Electronics Engineering | UPTU, Lucknow | 2009–13 |
GenAI Engineer · RAG Systems Architect · LLMOps · Agentic AI · Prompt Engineer
🏠 Remote-first preferred · Open to Hybrid (Lucknow / NCR) · India
Four production AI platforms. Zero budget. Real users. FastMCP & Agentic RAG.
© 2026 Ambuj Kumar Tripathi — All projects and certifications verifiable