I'm a backend and AI engineer who likes the unglamorous part of "AI engineering" — the queues, the retries, the schema migrations, the observability that lets an LLM pipeline run unattended at 3am without paging anyone. That's where most AI products actually die, and it's the part I've spent the last few years getting good at.

What I've worked on:

• 🏭 Enterprise AI enrichment — redesigned a classification pipeline on AWS Bedrock for UK manufacturing clients (Bentley, Heinz), improving resolution accuracy from 30% to 78.8%
• 🏥 Healthcare SaaS — architected Medicaid billing and EVV compliance automation across 19+ microservices, replacing multi-day manual billing cycles with scheduled batch jobs
• 🤖 Autonomous agents — built an LLM-agent job-application engine (Browser-Use + Playwright) that converts resumes into knowledge graphs and applies to jobs unattended
• 🎓 EdTech at scale — led a platform to 200,000+ users and 70M+ attempts while cutting infrastructure cost 50%

Right now: building and growing Examora — a test-prep platform for Pakistani engineering entrance exam candidates — where I own product, pricing, and content in addition to the code.

Currently exploring: multi-agent orchestration patterns, RAG evaluation methods, and GA4-driven growth analytics.

I enjoy problems where: a system has to reason (not just compute) and stay reliable under real production load — that intersection is where most of my recent work lives.

Languages

Backend

AI / LLMs

Frontend

Databases & Messaging

Cloud & DevOps

Automation & Testing

Test-prep platform for Pakistani engineering entrance exam candidates. Next.js Python PostgreSQL Growth/Content Systems

• Built and operate the full product end-to-end — engineering, pricing strategy, and content/growth
• Live product with active users — 🔗 examora.io

Autonomous agents that convert resumes into knowledge graphs and apply to jobs unattended. Browser-Use Playwright AWS Fargate SQS n8n

• Designed the queue-to-agent flow: SQS → Fargate workers → Aurora profile context → autonomous form completion
• Handled auth (2FA, OTP), CAPTCHA-solving, and missing-data fallback without unnecessary user prompts

Multi-pass LLM enrichment/classification system for enterprise ERP data. AWS Bedrock Amazon Neptune AppConfig

• Improved MFR resolution accuracy from 30.0% → 78.8% through prompt strategy and orchestration redesign
• Built versioned prompt/model config for safe, redeploy-free experimentation

Multi-tenant healthcare SaaS billing automation across 19+ microservices. Flask RabbitMQ Azure Service Bus ClaimMD API

• Converted a multi-day manual billing cycle into a scheduled automated batch job
• Auto-corrected EVV timestamp mismatches with full CMS audit logging

Textbook-grounded Q&A system for 200K+ students. Qdrant Mistral Vision Embeddings

• Extracted book content (including image context) and built similarity retrieval so students got answers grounded in real references

More on GitHub → repositories linked below once pinned (see recommendations at the end).

Ensemble.io — Backend & AI Engineer — Architecting AI enrichment pipelines on AWS Bedrock for enterprise manufacturing clients including Bentley and Heinz.

ALE Technologies — Backend & Cloud Engineer — Built multi-tenant healthcare SaaS automating Medicaid billing and EVV compliance for home health agencies.

Broomstick.AI — Full Stack & AI Automation Engineer — Built autonomous LLM-agent systems for web automation and job applications.

PreMed.PK — Technical Lead — Scaled an EdTech platform to 200,000+ users and 70M+ attempts, cutting infra cost 50%.

• 🔭 Growing Examora across product, content, and conversion
• 🧠 Exploring multi-agent orchestration patterns and RAG evaluation methodologies
• 📊 Building out GA4-driven funnel and conversion analytics tooling
• 🌱 Learning: advanced agent-reliability patterns (retry/backoff strategies for LLM-driven workflows)