📍 Karachi, Pakistan · Remote across US (EST/PST) and EU time zones · 📫 Open to AI Engineer roles
I build LLM systems that move real numbers, not demos. I took Refill Arena from 0 to $1M+ ARR and 100K+ monthly users, and I build pediatric health AI whose voice intake cut triage time 45% and whose RAG guardrails cut hallucinated recommendations 35%. I own the full lifecycle: retrieval, prompts, guardrails, deployment, and monitoring. I build in public as buildarc.ai on YouTube and Instagram.
| $1M+ ARR | 100K+ | 45% | 35% | 40% |
|---|---|---|---|---|
| Refill Arena, 0→1 | monthly active users | faster pediatric triage | fewer hallucinated recommendations | higher user engagement |
| Product | Outcome | What it is |
|---|---|---|
| Refill Arena | $1M+ ARR, 100K+ MAU, 3,200+ brands, 190+ countries, 95% uptime | Founding engineer on an e-commerce refill and subscription platform, built 0→1: backend infrastructure, data models, and scalable APIs |
| Little Angel Medical | 45% faster triage, 35% fewer hallucinated recommendations, fewer clinician escalations | Founding engineer and FDE leading development of a pediatric health AI platform across the US and Canada: computer vision and deep learning for at-home child health monitoring, a graph-based agentic memory over health records, and a real-time voice AI intake agent with RAG guardrails (citation checks, confidence thresholds, clinician fallback) |
| AI Wellness Platform + Virtual Trainer | 40% higher user engagement | LangChain + RAG pipelines on Apple HealthKit data that generate personalized nutrition and workout plans, plus an AI Virtual Trainer agent with agentic reasoning |
| Agentic Transcription Pipeline | Automated analysis that previously required manual review | AWS SQS to LLM agentic workflow that ingests 1hr+ BPO call transcriptions and runs analysis at scale |
Public builds where I push agentic AI, voice, and memory further.
| Project | What it does |
|---|---|
| memora-ai | Persistent memory layer that spans every AI tool you use (ChatGPT, Claude, Cursor), built on a hybrid graph-vector engine so your agents stop starting from zero |
| voker-agent-demo | Voice-agent observability: transcripts, failed tool calls, latency, handoffs, and likely root causes on one evidence-backed timeline, accessible over MCP |
| ai-meeting-assistant-app | AI meeting copilot with real-time transcription, streaming RAG chat, and automatic summaries and action items, running on a local SQLite vector store with no external database |
| family-care-agent | Turns family health records and voice notes into user-confirmed care plans, tracks medications, and sends WhatsApp reminders |
| ambra-revenue-guard | Learns from resolved insurance denials, stores the patterns in vector memory, and flags risky claims before you submit |
| opencal-agents | A calendar for your AI coding agents (Claude Code, Codex, Cursor): see every session on one timeline and schedule work by dropping tasks on the grid, the scheduler fires each job when capacity frees up |
| job-automation-n8n | Autonomous n8n + LLM pipeline that tailors a resume from a job description overnight, saving 90%+ of manual time |
| pixel-buddies | Animated desktop buddies that remind you to drink water, built with Electron |
| FileWheel | Fast, private, 100% local file converter and toolbox for macOS, Windows, and Linux |
Python · TypeScript · LangGraph · LangChain · RAG · MCP · Voice AI · FastAPI · NestJS · Node.js · PostgreSQL · Redis · Docker · AWS
LinkedIn · Portfolio · Instagram · YouTube · X · Email
If your team is hiring an AI engineer who ships outcomes end to end, my DMs are open.



