AI Engineer focused on LLM inference, serving infrastructure, and MLOps. I build and operate production LLM systems end to end — quantized model serving, inference optimization, RAG, guardrails, and full-stack observability.
- 🔭 Sole AI engineer at an early-stage startup — I own the AI/LLM stack from model serving to deployment, building production systems for regulated financial-services (fintech) use cases.
- ⚡ I like making models fast and cheap to run (quantization, replica pooling, latency/throughput tuning).
- 🎓 B.Tech, Electronics & Communication — IIIT Kottayam.
- 📫 Reach me: LinkedIn · satyamsingh20942@gmail.com
- LLM serving & inference — llama.cpp · LiteLLM · vLLM · GGUF quantization · on-demand replica pools · TTFT/throughput optimization
- RAG & evaluation — hybrid retrieval (FAISS + BM25 + RRF) · RAGAS · DeepEval · knowledge-graph vs. vector benchmarking
- LLMOps & observability — Langfuse · Prometheus / Grafana / Loki / Tempo / Pyroscope · prompt-injection guardrails
- Infrastructure — Azure · OpenTofu/Terraform · Ansible · Docker · CI/CD · Tailscale (zero-trust networking)
- 🧩 Open-source contributor to Apple MLX — 4 merged PRs improving benchmark timing reliability (
perf_counter) and documentation/API examples → view my commits - 📄 IEEE Published — Digital Twin for Drone Indoor Autonomous Navigation: GPS-denied indoor navigation with YOLOv11 + monocular depth estimation (Depth Anything), 0.994 mAP, 0.015 m RMS trajectory deviation → read the paper
llama.cpp · LiteLLM · vLLM · MLX · RAG · FAISS · Langfuse · RAGAS · Ansible

