I'm an AI Agent Engineer focused on building software that goes beyond generating answers — systems that understand context, make decisions, use tools, interact with external services, maintain state, and complete multi-step workflows.

My work sits at the intersection of AI agents, backend systems, developer tooling, APIs, automation, and full-stack product engineering.

I've built systems involving:

• 🤖 Autonomous and tool-using AI agents
• 🔌 MCP servers and agent-tool interfaces
• 🧠 LLM orchestration and structured outputs
• 🔄 Durable multi-step workflows
• 🛡️ Deterministic validation and safety boundaries
• 🔗 API, webhook, and third-party integrations
• 📚 RAG and retrieval-backed workflows
• ⚡ Event-driven automation
• 🧪 Agent verification and evaluation workflows
• 💳 Payment and transaction infrastructure

Most recently, I worked as an AI Automation Associate at Fiveral Ventures LLP, building LLM integrations, API-driven workflows, and event-based automation systems.

I enjoy turning ambiguous business problems into software that can reason, act, verify, and recover from failure.

An autonomous payment recovery system that detects failed payments, diagnoses the failure, chooses a recovery strategy, executes approved actions, and independently verifies the final outcome.

The system uses AI for reasoning while deterministic code controls financial execution and safety boundaries.

• Durable LangGraph workflows
• Real Razorpay API and webhook integration
• RAG-supported failure diagnosis
• Structured LLM outputs
• Human approval interrupts
• Deterministic recovery policies
• Idempotent external actions
• Independent provider verification
• Persistent audit and verification receipts

Tech: TypeScript • LangGraph • LangChain • OpenRouter • PostgreSQL • Supabase • pgvector • Razorpay • Zod • Vitest

An AI-guided Company Brain builder that turns a founder's tribal knowledge into structured SOPs, judgment rules, and skills that their team — and their AI agents — can use from day one.

A 5-question coaching flow compiles into KNOWLEDGE.md, PROCESSES.md, JUDGMENT.md, a shareable page, and a live JSON API. A LangGraph multi-agent pipeline (interviewer → completeness checker → writer) runs the coaching, RAG powers cited Q&A over every brain, and an eval harness guards quality in CI.

• LangGraph multi-agent coaching flow with Supabase checkpointing
• RAG with cited answers over generated company brains (pgvector)
• JSON-schema-constrained structured extraction with deterministic validation
• Anti-hallucination design — never invents unstated details
• Agent harness with dry-run mode and append-only audit log
• Eval harness (faithfulness, citation precision) gating CI
• Streaming coaching chat with Hinglish-aware prompting
• Live JSON API and shareable read-only brain pages

Tech: TypeScript • Next.js • LangGraph • Supabase • pgvector • OpenRouter • Tailwind CSS

A developer tool that converts vague software requirements into explicit, human-approved, testable behavioral contracts before an AI coding agent implements the wrong interpretation.

It combines an MCP server, Codex workflow, structured LLM analysis, deterministic validation, human approval, generated tests, and durable decision receipts.

• MCP server with multiple agent tools
• OpenAI Responses API integration
• Codex skill workflow
• Structured JSON Schema outputs
• Human-in-the-loop decision gates
• Deterministic semantic validation
• Test contract generation
• Vitest verification
• Durable decision receipts

Tech: TypeScript • MCP • OpenAI • Codex • TanStack Start • Zod • Vitest

A production-grade preflight scanner for agent-service endpoints — the preflight check for the agent economy. A LangGraph-orchestrated battery of deterministic network, protocol, and payment-challenge probes validates any endpoint and returns a readiness score with evidence for every check and RAG-grounded fixes for every failure.

• LangGraph-orchestrated audit pipeline (validate → safety gate → fetch → classify → aggregate → remediate)
• RAG-grounded remediation — fixes cite similar past failures instead of templated guesses
• Validated tool harness with timeouts and byte budgets
• Eval harness gating CI with a golden suite over adversarial cases
• Endpoint readiness validation
• x402 protocol inspection
• SSRF protection
• DNS and redirect validation
• Bounded network requests
• Deterministic scoring
• Structured evidence and remediation
• Production API endpoint

Tech: TypeScript • TanStack Start • React • LangGraph • LangChain • OpenRouter • Zod • Nitro • x402 • Bun

I'm currently going deeper into the engineering required to make autonomous agents reliable in production:

• 🤖 Autonomous tool-using agents
• 🔌 MCP servers and agent-tool protocols
• 🧠 Agent orchestration and durable execution
• 📊 Agent evaluation and observability
• 🛡️ Deterministic guardrails and verification
• 💾 Memory, RAG, and persistent agent state
• 🔄 Event-driven and long-running workflows
• ⚡ AI-native developer tooling
• 🌐 Production AI infrastructure

I'm particularly interested in systems where AI has to do more than generate text.

I like building agents that can:

Reason → choose tools → take actions → observe results → recover from failures → verify outcomes

The problems I find most interesting are where probabilistic AI meets deterministic software, external APIs, business rules, and real-world consequences.

I'm interested in working on:

• AI Agents
• Applied AI Products
• AI-Native Software
• Developer Tools
• MCP & Agent Infrastructure
• AI Automation
• Production LLM Systems
• Open Source

If you're building ambitious AI systems, I'd love to connect.