Software architect with 15 years in backend and platform engineering, now designing and running LLM agent systems in production: tool use through Model Context Protocol (MCP), unattended scheduled agents, and guardrails and cost control for AI coding agents.
Brings the platform discipline agents need to be trusted: service boundaries, idempotency, audit trails, rollback plans and specifications other engineers build from.
Owns the architecture of a 13-microservice commercial platform on Amazon Web Services (AWS) with TypeScript, Python and PostgreSQL, and leads a squad while guiding about 10 engineers.
Agentic AI: LLM agents, tool use / function calling, Model Context Protocol (MCP) servers and clients (stdio, Streamable HTTP), scheduled agent orchestration, human-in-the-loop review, guardrails, token and cost control, AI-assisted development (Claude Code)
LLM engineering: Retrieval-augmented generation (BM25 + Gemini embeddings, reciprocal rank fusion, cited sources), retrieval evaluation (hit@1 / hit@5 in CI), JSON-schema structured output (Zod, Gemini responseSchema), multi-provider integration (Anthropic Claude SDK, Google Gemini, local models via Ollama), context budgeting, image generation (Gemini image models)
Real-time AI (contributed): Voice agents on LiveKit/WebRTC, OpenAI and Gemini realtime models, speech-to-text build-vs-buy evaluation (managed APIs vs self-hosted Whisper)
Architecture: System and solution architecture, microservices, event-driven architecture, API gateway (JWT, rate limiting), backend-for-frontend, ports and adapters, idempotency, architecture decision records, build-vs-buy evaluation
Languages & Frameworks: TypeScript, Python, Node.js (Express, Hono), JavaScript, SQL, React, Angular, PHP (Laravel)
Data & Cloud: PostgreSQL, MySQL, Redis, MongoDB, Elasticsearch, ClickHouse, zero-downtime migration; AWS (S3, SQS, RDS, CloudFront), Azure, Docker, CI/CD
Leadership: Technical roadmaps, phased delivery, risk and rollback planning, stakeholder communication, mentoring senior engineers, Agile (SAFe, Scrum)
claude-toolkit — guardrail and token-cost tools for Claude Code agent sessions. github.com/hardyalsinghdeora/claude-toolkit
case-studies — public architecture decision record of the live migration. github.com/hardyalsinghdeora/case-studies
Multi-provider AI code reviewer (built, not yet launched) — Chrome extension reviewing Bitbucket pull requests through Anthropic Claude, Google Gemini or a local Ollama model behind one interface; schema-validated findings anchored to diff lines; noise-filtered, budgeted diffs. TypeScript, React.
Image-generation MCP server — MCP server giving Claude image generation and editing via Gemini image models; one tool module behind stdio and Streamable HTTP transports; keychain-held API key; container-ready for Cloud Run. Node.js.
Portfolio agent with hybrid RAG — Gemini function-calling agent on Cloudflare Workers that retrieves from a chunked career profile (BM25 + embeddings, reciprocal rank fusion) and cites its sources; retrieval eval set included. github.com/hardyalsinghdeora/portfolio
Ops Copilot (learning project) — LangGraph multi-agent RAG over MCP servers, LLM-as-judge review, human approval via interrupt(). github.com/hardyalsinghdeora/ops-copilot
B.Tech, Electronics and Communications Engineering, Rajasthan Technical University (RTU), Kota — 2011
Certified SAFe 5 Agilist, Scaled Agile — 2022
Google Project Management Certificate, Google / Coursera — 2022