Nadeem Ahmad

I'm a Machine Learning Engineer and Full-Stack AI Developer with 1.5 years of experience shipping production-grade AI systems — autonomous agents, multimodal pipelines, and enterprise RAG — not demos or API wrappers.

I specialize in the hard part: taking AI from research to reliable, cost-efficient production. That means deterministic agent orchestration that doesn't loop infinitely, retrieval pipelines that don't hallucinate, and inference backends that don't collapse under load. On the web side, I deliver complete end-to-end systems — async FastAPI backends wired to Next.js frontends with full Docker containerization and CI/CD.

What I Build

Autonomous Multi-Agent Control Planes Stateful agent orchestration using LangGraph with deterministic state machines, Redis event-sourced audit logs, circuit breakers, and dead-letter queues. Production agents that recover from failures gracefully instead of spinning in infinite loops.

High-Throughput LLM Inference Custom vLLM serving with PagedAttention virtual memory management and CUDA kernel optimization for sub-second, high-concurrency token generation — built for real traffic, not toy benchmarks.

Enterprise Hybrid RAG Pipelines Sparse BM25 + dense vector retrieval fused with LlamaIndex, Cross-Encoder rerankers, and property graph indexes. Multi-stage hallucination verification layers for clinical and enterprise-grade accuracy.

Universal AI Gateways LiteLLM proxy routing across 100+ LLM providers with virtual tenant API keys, semantic caching, automated failovers (OpenAI → Claude → Bedrock), and sub-8ms P95 latency at scale.

Self-Improving Prompt Compilers DSPy declarative signatures with MIPROv2 Bayesian prompt optimization — eliminating brittle, hand-written prompt strings and replacing them with compiled, type-safe LLM programs.

Recent Production Work

Built an autonomous AI voice calling agent for SMB clients using n8n, Twilio, and the OpenAI Realtime API — automating inbound inquiries and appointment booking end-to-end with low-latency real-time orchestration.

Engineered multi-agent RAG pipelines and specialized multimodal fine-tuning pipelines (LoRA/QLoRA) on domain-specific transformer models to improve output quality and reduce inference cost.

Tech Stack Python · PyTorch · vLLM · CUDA · LangGraph · LiteLLM · DSPy · LlamaIndex · FastAPI · AsyncIO · Next.js 14 · TypeScript · Redis · PostgreSQL · Docker · OpenTelemetry · Twilio · n8n

Currently ML Engineer at NeuroWebLabs and BS Software Engineering student at FAST NUCES, Pakistan. I work async-first, communicate clearly, and take full ownership from architecture through deployment. Open to remote ML Engineering and Full-Stack AI roles globally — actively targeting Germany and international markets.