Summary
Founder and Full-Stack AI Engineer with deep expertise in RAG systems, serverless AI, vLLM, model hosting, and MLOps. Built KRAG, a serverless RAG platform that beats Google NotebookLM in PDF parsing and table extraction. It runs on scale-to-zero GPUs so you pay nothing when idle. Strong in Vector DBs (BGE-M3), MCP, Modal, and bridging backend (PostgreSQL, Redis, tRPC) with AI tooling (Vercel AI SDK, fine-tuning). Track record: zero-downtime migrations, 90% RAG accuracy gains at Buildway.ai. Now leading SMAKG and building flagship products (KRAG, The Informant) plus client work like StealthNode.
Core Skills
- AI & RAG: Vercel AI SDK, MCP (Model Context Protocol), Vector DBs & Embeddings (BGE-M3), RAG Optimization & Citation, LangChain, vLLM & LLM Orchestration, Modal (Serverless GPU), Fine-tuning, Puppeteer (Automation).
- Frontend: React, Next.js 14+, TypeScript, Tailwind CSS, Recharts, Shadcn UI, Responsive & SSR.
- Backend: Node.js, tRPC, Prisma, PostgreSQL (Neon / Supabase), Redis, Better Auth, NextAuth, Zod.
- DevOps & Tools: Docker, Kubernetes, CI/CD (GitHub Actions), AWS, Modal (Serverless), Vercel, E2B (Sandboxing), Cloudflare Tunnel.
Professional Experience
- Company Building: Founding and leading SMAKG.com; defining product vision, technical strategy, and go-to-market.
- Product & Engineering: Driving development of flagship products (KRAG, The Informant) and core infrastructure.
- Client Development: Building client products under SMAKG (e.g. StealthNode, an AI-driven SOC platform).
- AI Infrastructure & RAG Optimization: Engineered a multi-agent system with Dynamic Chain of Thought and Web Search tools using MCP. Improved RAG citation accuracy by 90% and data retrieval by 70% via complex vector DB logic and post-hoc processing.
- Solo Lead (Billing & AI): Built complex billing infrastructure and shipped a full AI chat system with dynamic interactive charts (Vercel AI SDK, Recharts) in <2.5 weeks. Debugged critical production "ghost migrations" with 100% reliability.
- Critical Zero-Downtime Migration: Executed a high-stakes migration from NextAuth to Better Auth for a live production app, resulting in 0% downtime and zero data loss.
- Advanced Security & Infrastructure: Refactored tRPC infra for security (zip-bomb protection, E2B sandboxing). Implemented 100% secure Redis-backed pagination for MCP, boosting client agent accuracy by 90%.
- High-Velocity Engineering: Contributed to 3 production projects from Day 1 (zero onboarding). Parsed complex Austrian legal APIs (big data seeding, complex joins) and reduced launch times by optimizing DB load and caching.
Products launched
- Beats Google NotebookLM in PDF parsing and table extraction (March 2026). Architected the world's first serverless RAG agent: Next.js API plus Modal Python workers and Redis task queues. 40% lower cost than major parsers while preserving tables, structure, and formatting.
- Serverless GPUs scale to zero. No idle costs; scales by design. Multi-format ingestion (PDFs, web URLs) with 3 enterprise encryption levels, parent-child chunking, real-time status, and context-aware chat with citations.
- Florence-2 for image understanding and search. Custom models on Modal: Marker PDF, BGE-M3, MXBAI Reranker, Qwen 2.5 14B. All serverless with auto-scaling.
- Stack: Next.js 16, tRPC, Prisma, Redis, Supabase, Exa, Modal, vLLM, model hosting, MLOps.
- AI-powered Security Operations Center built by SMAKG for a client. Deploys via lightweight MCP + agent installer; logs stream to Wazuh; custom rulesets trigger an AI agent in isolated Modal sandbox connected via secure Cloudflare MCP tunnel.
- Agent detects and neutralizes threats in real time; delivers detailed incident reports. Stack: Modal, MCP, Wazuh, Cloudflare Tunnel, event-driven architecture.
- Custom fine-tuned SLM for CS students. [PROSE] explains concepts like a 1940s noir detective; [CODE] returns ready-to-use Python code.
- Serverless deployment on Modal; actively under development. Stack: Modal, Python, SLM, fine-tuning.
- Architected and launched a multi-tenant SaaS platform serving 50+ active users, featuring custom domain support (Vercel-style routing) and dynamic subdomains via Next.js Middleware.
- Integrated an AI Assistant to automate user support and enhance portfolio creation workflows.
- Implemented robust CI/CD pipelines using GitHub Actions for dev and prod branches, automating linting, testing, and builds to prevent regressions.
- Enforced end-to-end type safety using Zod across all API endpoints and database schemas.
Open Source Contributions
- Merged 8+ PRs in a single month, working alongside core senior developers on high-priority tickets and received direct mentorship.
- Contributed a significant portion to the new calendar-based Kanban feature, implementing complex frontend logic and interactions.
- Integrated Vitest for type-safe unit testing and resolved critical UI formatting bugs affecting thousands of users.
Freelance Experience
- Delivered a secure, containerized (Docker) web application using Next.js and MongoDB.
- Built a reusable UI component library and containerized the app to reduce deployment friction and ensure environment consistency.
Education
Bachelor of Science in Data Science, Indian Institute of Technology, Madras
Exp. Jan 2029
Certifications
Docker Certification – Docker, Inc.
2025
GitHub Actions Workshop: CI/CD Pipelines – Microsoft Press
2025