Hi, my name is

Rachana Shree H S.

I build agentic AI systems.

I'm an AI/ML Engineer shipping production agentic and RAG systems on Azure for healthcare, and I measure model quality before anything goes out the door.

View Projects Get In Touch

01. About Me

Hello! I'm Rachana, an AI/ML Engineer with 2.3 years of experience building production Agentic AI, RAG and LLM-based systems. I have a strong background in hybrid retrieval, model evaluation, FastAPI, ASP.NET Core and Azure, and I measure model quality before shipping.

At Gainwell Technologies I build LangGraph workflows on Azure for healthcare guideline analysis, relationship-aware retrieval, human-in-the-loop approvals and claim fraud detection. Alongside that I build projects in hybrid retrieval, LLM fine-tuning, verified code generation and agri-advisory, including Kisaan Sampurna, now at 50,000+ installs on Google Play.

My project work has received KSCST Karnataka state funding and a 2nd place at the IISc Bangalore Research & Innovation Program.

Technologies I've worked with recently:

  • LangGraph
  • LangChain
  • CrewAI
  • Azure OpenAI
  • Claude
  • FastAPI
  • ASP.NET Core
  • PyTorch
  • QLoRA
  • MLflow
  • Azure ML
  • React
RS
Rachana Shree H S

02. Where I've Worked

Associate AI/ML Engineer @ Gainwell Technologies

Jan 2025 – Present · Bengaluru

  • Built agentic workflows (LangChain, LangGraph, FastAPI, Azure OpenAI, Claude) for healthcare guideline analysis and grounded response generation, serving 40+ users and 500+ requests/week.
  • Implemented RAG over Azure AI Search and Cosmos DB Gremlin for relationship-aware retrieval. Added PostgreSQL-backed LangGraph checkpointing for fault recovery and human-in-the-loop approvals.
  • Automated RFP response with a CrewAI and Azure Document Intelligence pipeline for requirement extraction and draft generation, cutting first-draft time from 8 to 2 hours.
  • Built a claim fraud detection pipeline (Azure ML, MLflow, Scikit-Learn) with tracked experiments and ONNX serving for real-time and batch inference. It reached 0.93 AUC and 82% precision at 70% recall on 250K claims.
  • Developed LLM-powered engineering analytics on Azure DevOps APIs and Bronze/Silver/Gold claims-reconciliation workflows, reducing manual reconciliation effort by 40%.

03. Things I've Built

Featured Project

Sentinel

Agentic Financial Research & Regulatory Compliance Engine

Fine-tuned Qwen-2.5-7B Instruct with 4-bit QLoRA on 2,174 domain examples. On 271 held-out test claims it reached 0.996 Macro F1 and 99.6% accuracy (95% CI 98.0–99.9%), a +39.5% relative lift over GPT-4o-mini. Observability, evaluation and QLoRA pipelines with Langfuse improve retrieval quality, financial reasoning and the reliability of AI-generated research reports.

  • 0.996Macro F1
  • +39.5%vs GPT-4o-mini
  • 9xcost cut · $1 vs $9 / 1k
  • 612 msp50 latency
  • LangGraph
  • QLoRA
  • Qwen 2.5
  • BAAI/bge-reranker-base
  • RAGAS
  • FAISS
  • PostgreSQL
  • FastAPI
  • Langfuse
  • Docker
  • AWS

Featured Project

Legacy Modernization Copilot

AI-Powered .NET Refactoring

A .NET modernization copilot built on Roslyn, hybrid RAG and Gemini. It automates identification of 18+ legacy code anti-patterns across repository-scale codebases. Repository-aware hybrid RAG improved retrieval quality by 32% over semantic search, and AI-generated refactors passed verification 78% of the time across 40 legacy .NET files.

  • 18+anti-patterns found
  • +32%retrieval vs semantic
  • 78%refactors verified
  • 40legacy .NET files
  • C#
  • Python
  • Roslyn
  • Gemini
  • Hybrid RAG
  • React
  • SQLite
  • Render
  • GitHub Actions

Featured Project

Kisaan Sampurna

Multimodal Agri-Advisory & Market Intelligence

A multimodal agri-advisory platform built with LangGraph, Neo4j and FastAPI, integrating a 3,140-node crop-disease knowledge graph and a market intelligence engine. Qwen-2.5 was fine-tuned on 178K+ training pairs, and ensemble forecasting models cover 6,273 mandis with 9.19% validation MAPE.

  • 50,000+installs
  • 3,500+subscribers
  • 6,273mandis forecast
  • 9.19%validation MAPE
  • Gemini 2.5
  • Qwen-2.5
  • LangGraph
  • QLoRA
  • Neo4j
  • GraphQL
  • FastAPI
  • XGBoost
  • PostgreSQL
  • AWS

04. AI Playground

Pick a project and run it. This is a simulated run that streams illustrative pipeline steps for each project. It is not live model output.

rachana@playground ~ zsh

05. Technical Skills

GenAI / LLMs

Agentic AI, RAG, Hybrid Retrieval, Reranking, Prompt Engineering, QLoRA Fine-Tuning, Model Evaluation, Responsible AI, Azure OpenAI, Gemini, Claude, Qwen

AI Frameworks

LangChain, LangGraph, CrewAI, FAISS, BM25, PyTorch, Scikit-Learn, TensorFlow

MLOps

Azure ML, MLflow, ONNX, Experiment Tracking, Model Registry, Model Serving, Langfuse

Data & Search

Azure AI Search, PostgreSQL, Cosmos DB (Gremlin), Neo4j, SQLite, Redis, GraphQL, Pandas

Backend & Cloud

Python, C#, SQL, FastAPI, ASP.NET Core, Microservices, Docker, Kubernetes, AWS, Azure, GitHub Actions

Frontend

React, TypeScript, JavaScript, Vite

Education

B.E., M. S. Ramaiah Institute of Technology

CGPA 9.32 / 10 · 2021 – 2025

Certifications

  • Building with the Claude API · Anthropic Academy
  • Generative AI Fundamentals · Databricks, Oct 2026

06. What's Next?

Get In Touch

For roles, collaborations, or questions about any of this work, my inbox is open.

Say Hello