Abhay M.

Full-stack AI engineer building AI agents, RAG chatbots, and LLM-powered SaaS that hold up in production. I work in Claude Code, Python, Next.js, and Supabase, and I take AI-built or half-built apps from demo to launch. Top Rated Plus with 90+ completed Upwork jobs. What I build: - AI agents and agentic workflows: tool calling, MCP servers, human-in-the-loop approvals, and multi-step automations that do real work - RAG chatbots and AI assistants over your PDFs, docs, knowledge base, website, or product data (pgvector, embeddings, LangChain, LangGraph) - LLM integration into existing products with the Claude API, OpenAI API, and Gemini: structured outputs, evals, and guardrails - AI SaaS MVPs and launches: Next.js, React, TypeScript, Node.js, Supabase, PostgreSQL, Stripe billing, Vercel deploys - API integrations and AI automation: CRMs, Stripe, Slack, Google Workspace, WhatsApp, webhooks, n8n, Make, and Zapier - Internal tools and dashboards that pull live data from the systems you already use - Finishing and fixing vibe-coded apps from Lovable, Bolt, Replit, Cursor, or Claude Code - LLM cost and latency optimization: model routing, caching, batching, and prompt tuning Security reviews for AI-built apps: about to launch, take payments, or raise on a Lovable, Bolt, Replit, Cursor, or Claude Code app? I find and fix what AI tools leave behind: Supabase tables anyone can read, exposed API keys, broken login and role checks, unprotected API routes, bypassable Stripe webhooks, and prompt injection in AI features. You get a severity-ranked report in plain English, fixes as PRs, and a re-test. How I work: AI-native. Claude Code, Codex, and Cursor do much of the typing. I own the engineering around them: written specs, reviewed diffs, tests, CI gates, and documentation. You get clean pull requests in your GitHub repo, not a pile of generated code, and a handover your team can maintain. Stack: Python, TypeScript, JavaScript, Node.js, Next.js, React, Tailwind, shadcn/ui, Supabase, PostgreSQL, pgvector, Drizzle, Prisma, Stripe, Vercel, Railway, Docker, AWS, GitHub Actions, Sentry, PostHog, Vercel AI SDK, LangChain, LangGraph, n8n Models: Claude, GPT, Gemini, DeepSeek, open models on Ollama and vLLM Past work includes an AI assistant for a US university, a college counselling chatbot, a function-calling CRM assistant, a second-brain tool for criminal lawyers, healthcare AI, a Stripe Connect marketplace, and an AI yoga and meditation guide. I overlap 4+ hours with US Eastern time, send short written updates, and will tell you when something shouldn't be built. Send me what you have and what's blocking you, and I'll reply with a plan.