Ganesh Ranganath Chandrasekar
Research Software Engineer · LLM pipelines & workflow automation
Reach me by email at ganesh@cgrx.dev, or connect on LinkedIn.
Summary
Engineer specializing in building LLM-powered pipelines for workflow automation across multiple domains. Experienced in taking projects from concept to production and comfortable across the full stack: model fine-tuning, inference serving, orchestration, and cloud deployment.
Education
M.S., Computational Science
University of California, San Diego · Sep 2016 – Dec 2018
B.E., Electrical and Electronics Engineering
Sri Venkateswara College of Engineering, Chennai · Aug 2011 – May 2015
Tech Stack
| Category | Tools |
|---|---|
| Core ML & Data | PyTorch, PaddlePaddle, NumPy, Pandas, Transformers, SpaCy, Docling, FasterWhisper |
| LLM Orchestration & Harness | LangChain, LangGraph, Hermes, Claude Code |
| Inference & Serving | Triton Inference Server, vLLM, Text Generation Inference, Text Embedding Inference |
| Retrieval & Vector Stores | FAISS, Milvus, Chroma, Redis |
| Ops & Infra | Docker, GitHub Actions, EC2, S3, Lambda, Fargate, API Gateway, CloudFormation, Secrets Manager |
| APIs & Data Access | OpenAI, TogetherAI, Playwright, Firecrawl, ZenRows |
| Backend & UI | FastAPI, Flask, OpenWebUI, Streamlit, Gradio |
| Evaluation & Observability | MLflow, WandB, LangFuse, LangSmith, Ragas |
Experience
Research Software Engineer
Santa Fe Research Pvt Ltd, IIT Madras Research Park, India · Nov 2021 – Present
- Built a RAG pipeline for contextual Q&A over multi-state healthcare policy documents using OpenWebUI, LangGraph, Docling, Chroma, and Langfuse.
- Developed an ASR pipeline with Whisper for call transcription, sentiment tracking, and automated post-call summarization.
- Built a LangGraph agent for dependency-aware cascade analysis to generate ordered modification sequences for complex executive travel itineraries.
- Designed a DAG to scrape news and generate per-entity risk profiles, processing 200+ articles in 5 minutes using Sentence Transformers and LLaMA 3.1.
- Fine-tuned YOLOv9 with Azure OCR for PAN card field extraction, achieving 93.7% accuracy and reducing KYC QA TAT and cost.
- Engineered a top-5 MCC recommendation system from store images, resolving annotation bottlenecks across ~250 codes to streamline client on-boarding.
- Designed a 3-layer LangChain architecture using ChatGPT and LLaMA2 for schema-constrained MCQ, fill-in-the-blank, and descriptive Q&A generation.
- Re-architected the inference stack to scale attribute detection from 100 to 15,000 images/hr, and fine-tuned metric-learning and CLIP models for semantic search.
Research Assistant
Rady School of Management, University of California, San Diego, USA · May 2019 – Jan 2020
- Generated visualizations of US Gun Crimes for geophysical analysis using ESRI tools.
- Developed R scripts for address sanitization, duplicate license identification, and name ambiguity resolutions.
Graduate Teaching Assistant
Department of Mathematics, University of California, San Diego, USA · Sep 2017 – Dec 2018
- Assisted 150+ students in Linear Algebra, Vector Calculus, and Differential Equations.
- Consistently achieved an "excellent" instructor rating with a 91% average student recommendation.