Agentic Systems · LLM Infrastructure · MLOps
Ahmedabad, India · darshan105panchal@gmail.com · LinkedIn · GitHub · Portfolio

AI & ML Engineer with 2+ years in production. I build and ship multi-agent LLM systems, agentic RAG pipelines, and inference-optimized backends — deployed across legal, space, pharmaceutical, and maritime domains using LangGraph, FastAPI, and Kubernetes. Reduced enterprise LLM inference costs by 60% via vLLM migration. Currently mentoring practitioners in deep learning and production LLM systems.

AI Agents & LLMs

Deep Learning & ML

Infrastructure & MLOps

Data & Vector Stores

Evaluation & Safety

LangGraph · FastAPI · Wikidata SPARQL · CourtListener

4-node agentic pipeline that extracts anatomical ontology IDs via Wikidata SPARQL, retrieves US case-law precedents via CourtListener, and synthesizes structured malpractice liability reports in seconds. · Live Demo

LangGraph · SGP4 · Groq · RunPod Serverless · CelesTrak

Multi-agent system fetching live TLE satellite data from CelesTrak, propagating orbital paths with SGP4, and applying maritime law analogies to adjudicate collision liability under the 1972 Outer Space Liability Convention. · Live Demo

LangGraph · FastMCP · OpenFDA · RunPod

Agentic pharmacovigilance system mapping social-media drug slang to MedDRA Preferred Terms and validating adverse drug reaction signals against FDA FAERS in real time. Deployed on RunPod Serverless. · Live Demo

LangGraph · FastAPI · Docker · RunPod · OBIS · Shapely

3-agent LangGraph pipeline (Navigator, Marine Biologist, Risk Manager) querying live OBIS cetacean sighting data and applying a 0–100 composite scoring model to select the safest, most efficient shipping route. · Live Demo

AI/ML Developer — NexusLink Services India Pvt. Ltd. Apr 2024 – Present

• Architected production LLM wrappers and autonomous RAG agents (LangGraph + Qdrant); migrated clients from proprietary APIs to self-hosted vLLM clusters on Kubernetes — 60% inference cost reduction
• Designed multimodal enterprise document ingestion pipelines with hybrid embeddings for citation-backed Q&A at scale
• Built FastAPI microservices and optimized ML data pipelines — 40% processing time reduction
• Fine-tuned classification, regression, and clustering models (PyTorch, scikit-learn, XGBoost) — 30% accuracy improvement

Deep Learning Mentor (1:1) — Independent · Remote 2026 – Present

• Structured mentorship covering neural networks, CNNs, transformers, and production LLM deployment
• Guiding mentees through hands-on model training, optimization, and cloud deployment workflows

• M.E. Transportation Engineering - L.D. College of Engineering 2020 – 2022
• B.E. Civil Engineering - Gandhinagar University 2016 – 2020
• Data Science Pro Program - Fingertips Academy Sep 2023 – Mar 2024