About Me
AI engineer with 3+ years building production LLM agent systems in Python: tool-calling agent loops, custom orchestration, context management, structured outputs, and retrieval-augmented pipelines. Hands-on across the Op…
AI engineer with 3+ years building production LLM agent systems in Python: tool-calling agent loops, custom orchestration, context management, structured outputs, and retrieval-augmented pipelines. Hands-on across the OpenAI Agents SDK, Anthropic Claude API, Model Context Protocol, LangChain and LangGraph, FastAPI backends, and PostgreSQL with pgvector. Designs agents for environments where correctness is auditable: deterministic routing instead of LLM routing where every action must be traceable, scoped tool access, single-mutator write discipline, guardrails on untrusted input, and human-in-the-loop approval on sensitive actions. Comfortable operating end-to-end across the agent stack — architecture, build, deploy, and post-launch debugging — and speaking concretely to failure modes and the mitigations that fixed them. Has enabled 2,000+ engineers and 100+ faculty on agent engineering and MCP.
Experience
Lead AI Engineering Instructor
<p>Owned end-to-end delivery of an applied AI engineering curriculum covering Python, the OpenAI Agents SDK, multi-agent orchestration, tool calling, RAG pipelines, Model Context Protocol, and REST API integration, delivered to 2,000+ engineers across 24 cohorts.</p><p> Built production-grade Python reference implementations and technical modules reused across 8 courses, establishing agent architecture standards and engineering practice across the programme.</p><p> Acted as the technical escalation point for 100+ faculty on LLM application development, MCP, multi-agent orchestration, and prompt engineering — diagnosing agent failure modes in other people’s codebases under time pressure.</p><p> Defined and validated 10 AI engineering competencies with Government of Sindh officials and 6 AI specialists, translating architecture decisions for non-technical executive stakeholders.</p><p> Ran hands-on live build sessions, debugging unfamiliar codebases and deployment environments in real time in front of an audience.</p><p> Progressed from Teaching Assistant to Co-Lead to Lead Instructor; selected by programme leadership to lead a flagship cohort.</p><p> Recognised by the Government of Sindh.</p>
AI Engineer, Multi-Agent Systems
<p>Architected and deployed multi-agent AI systems on the OpenAI Agents SDK, taking automation workflows from requirements gathering through production deployment.</p><p> Built and maintained Python backend services and integrated LLM-powered tooling with external REST APIs, owning integration design, error handling, retry logic, and post-launch reliability.</p><p> Designed and shipped conversational AI interfaces adopted by cross-functional teams for real-time collaboration.</p><p> Worked embedded with engineers and product managers across the full delivery lifecycle, translating ambiguous business requirements into shipped, supported systems.</p>
PROJECTS
ResolveAI — Bilingual Voice-and-Text Banking Resolution Agent
<p>A production conversational AI agent handling banking complaints in Urdu and English, over both voice and text. Intent classification feeds a deterministic Python routing layer — chosen deliberately over LLM-based routing so every action taken is reproducible and auditable. The tool-calling layer uses scoped access per intent with structured outputs validated before any write, so a malformed or adversarial model response cannot reach the database. Cross-session memory via pgvector similarity search means returning users don't restate context. Built under enterprise governance constraints: single-mutator write discipline, SLA state machine, full audit logging, no sensitive user data exposed to the model provider, and human-in-the-loop escalation with zero automated money movement by design. Built solo, end-to-end — architecture, backend, agent layer, voice pipeline, deployment. Stack: Python, FastAPI, OpenAI Agents SDK, PostgreSQL/pgvector, Whisper STT, OpenAI TTS, Docker</p>
Multi-Agent Coaching System — WhatsApp Delivery
<p>A multi-agent system delivered over WhatsApp rather than a standalone app, so users adopt it inside an existing daily habit — a deployment-first architecture decision that removed the install barrier entirely. I built the agent orchestration, persistence, and scheduling layers, and operated the deployed service on Railway with managed PostgreSQL on Neon. Stack: Python, OpenAI Agents SDK, FastAPI, PostgreSQL (Neon), Railway, WhatsApp Business API.</p>