Arsalan K.

I build AI products that work in production, not just in demos. I'm a Full-Stack AI Engineer with 3+ years shipping LLM apps, RAG pipelines and AI agents on Python backends (FastAPI, Django), with React/Next.js frontends and AWS, GCP and Azure deployment. What I've shipped recently: • CareCompass AI (US healthcare, HIPAA-compliant): a LangGraph text-to-SQL agent on BigQuery with RBAC filter injection and SQL validation; an AI document-review pipeline that reads scanned faxes with Claude, Gemini and GPT-5 and applies 19 clinical decision rules; an async job system with checkpoints and retries; and an LLM evaluation gate that runs a golden dataset on every pull request. • BlackGrid AI CFO (fintech SaaS): Django REST backend and an AI CFO on the OpenAI Responses API with tool calling and cited web search, integrated with Plaid, QuickBooks, Gmail and Stripe, shipped to web, iOS and Android. • TradieMate (SaaS for electrical contractors): a provider-agnostic LLM layer (OpenAI, Anthropic, Gemini), job-level RAG chat on pgvector with streaming, and AI quote review. • FreshPass AI (salon SaaS): sole backend developer, building a real-time AI voice receptionist with function calling, multi-tenant RAG assistants and an AI hairstyle try-on pipeline. • Legal research platform: grounded RAG chat with inline citations, and AI document drafting streamed into a React editor. What I do: • LLM integration: OpenAI, Claude, Gemini, structured outputs, tool calling, multi-provider fallbacks • RAG and vector search: pgvector, Pinecone, Qdrant, ChromaDB, with citations and evaluation • AI agents: LangChain, LangGraph, DSPy, MCP, voice agents • Backend and APIs: FastAPI, Django, Celery, Redis, PostgreSQL, webhooks, streaming • Full-stack and deployment: React, Next.js, TypeScript, Docker, AWS, GCP, CI/CD How I work: I measure AI quality with evaluation sets instead of assuming it, build error handling and retries in from day one, ship in small reviewed pull requests, and send clear written updates. Tell me what you're stuck on, and I'll tell you how I'd fix it, or tell you if it isn't worth building.