Backend and AI engineer. I build systems end to end and keep them running.
orqestra.xyz · manavvgarg.github.io · LinkedIn · manav.garg@hotmail.com
Production backend services in Python, Go and Java, deployed on AWS with Docker and Kubernetes. LLM systems that have to hold up under real use rather than demo well: retrieval pipelines, agent orchestration, evaluation frameworks, and the guardrails that keep output trustworthy.
Most of my time goes on the half people skip. Reliability, evals, and figuring out why something broke at 2am.
Currently: building Orqestra, a self hosted agent orchestration platform. Multi step workflows with tool calling, sandboxed execution, full run trajectory capture, and eval tooling that replays runs across model and prompt versions.
Orqestra — Self hosted AI orchestration platform. Go, Python, Docker, Redis. Built solo, live and public.
HoneyML — LLM driven adaptive SSH honeypot. Generates a coherent shell per attacker session and maps observed behaviour to MITRE ATT&CK.
LinkedIn Profile API — Structured profile data through an undocumented internal API. Runtime endpoint version discovery, explicit schema drift errors instead of plausible wrong data.
Vision Transformers — ViT, DeepViT and Compact Convolutional Transformers benchmarked on medical and satellite imagery. 95%+ on skin lesion classification, 96% on land use.
Languages Python · Go · Java · TypeScript · SQL · C++
Backend FastAPI · Django · Node.js · REST · WebSockets · async and event driven services
AI/ML PyTorch · TensorFlow · HuggingFace · LangChain · MCP · RAG · embeddings and vector search · LoRA/PEFT · DeepSpeed ZeRO
Data PostgreSQL · MongoDB · Redis · Spark · Kafka · Airflow · ETL and data lake architecture
Infra AWS · Docker · Kubernetes · CI/CD · Linux
I write about inference optimisation, agent architectures and LLM failure modes on LinkedIn.