Hey there! I'm Saish Shetty

Data Scientist & AI/ML Engineer

Passionate about Deep Learning—especially Generative AI, CNNs, and Transformers. I love exploring Machine Learning, Data Science, and Software Development. Currently working on Healthcare Fraud Detection & Risk Scoring.

Saish Shetty

About Me

A passionate Data Scientist and AI Engineer with expertise in Generative AI, Large Multimodal Models, and Agentic Systems. Currently working at Bajaj Finserv Health Limited, I've built AI solutions that have significantly improved business metrics.

AI Innovation

Built Enigma, a GenAI platform reducing claims processing time by 50%

Recognition

Sharp Shark Award & PropelX 4.0 program selection

Education

B.Tech in Information Technology from Manipal Institute of Technology (CGPA: 8.29)

Fun fact: I can train a neural network all night… but I'll never skip a dance break! 💃

2+ Years Experience
10+ AI Projects
3 Awards

Experience

BFHL Logo
Bajaj Finserv Health Jan 2023 - Present · Pune, India

Data Scientist

Nov 2024 - Present

Fraud Detection & Risk Scoring

Building an end-to-end tree-based fraud prediction model with 200+ temporal, behavioral, and clinical features achieving PR-AUC of 0.454 on highly imbalanced real-world claims data. Developed an explainability framework combining SHAP-driven insights with historical pattern analysis.

Straight Through Processing (STP)

Built Enigma, a GenAI claims platform modernizing legacy adjudication; reduced AHT by 50% and AER by 37%. Designed SOP compliance pipeline on Large Multimodal Models driving 50% MoM increase in auto-approvals. Today, 100% of claims flow through Enigma with fault-tolerant microservices using Apache Kafka.

Associate Data Scientist

Jul 2023 - Oct 2024

Customer Service Automation

Built an Email and App Query automation service using hybrid AI models, increasing Non-Voice Auto-Reply Rate from 3% to 40%.

ABHA Onboarding Project

Engineered a distributed facility intelligence pipeline on the National Healthcare Providers Registry using Python on AKS containers, uncovering ~80% previously unmapped facilities for market research and expansion.

Security Compliance Bot

Developed using Azure OpenAI (GPT-4), FastAPI and Chroma vector DB to automate vendor security questionnaires with structured answers and evidence links, reducing manual effort for third-party assessments.

Data Science Intern

Jan 2023 - Jun 2023

HealthSaathi

Built a full-stack Agentic AI chatbot using Retrieval Augmented Generation (RAG) with Langchain and OpenAI's GPT-3.5. Created multiple tools for task execution and enabled spin-up of ailment-specific chatbots with context-aware system prompts.

Pincode Serviceability Model

Designed a geospatial model using K-Nearest Neighbors and distance-weighted features to identify underserved regions, prioritizing high-impact pincodes for doctor onboarding and cashless coverage growth.

Tata CLiQ Logo
Tata CLiQ May 2022 - Jul 2022 · Mumbai, India

Data Science Summer Intern

May 2022 - Jul 2022

Metadata Search Interface

Built an internal metadata search interface integrated with AWS Redshift's metadata framework to improve schema discoverability across the data lake.

Metadata Management Roadmap

Contributed to a roadmap for scalable metadata cataloging and governance across the organization.

Featured Projects

ShadowBot

Everything AI

Full-Stack AI Overlay Desktop Application for Interviews & Sales Calls

Electron Gemini MCP Seraphine
  • Custom LLM-Orchestrator Backend with unique Multi-Agent DAG created as per user query
  • Persona-specific specialization coordinated via a structured graph and unified context management within a session.
  • Multi-MCP (Model Context Protocol) server aggregation
  • Powered by Google's Gemini AI model, with runtime Gemini API Key injection

Navigators.AI

Agentic AI Travel Application

An Agentic AI Travel Application that plans the best itineraries from multimodal user input

Llama-8B CLIP AWS ECS Docker
  • Used a custom fine-tuned Llama-8B model for itinerary planning and functional tool calling
  • Integrated real-time flight prices and hotel prices fetching
  • Implemented custom trained CLIP model for geo-location identification
  • Deployed on AWS ECS with multiple docker containers

ResNet-50 on ImageNet-1K

Mixed-precision DDP Training

Mixed-precision DDP training of ResNet-50 from scratch on AWS EC2 using PyTorch

PyTorch AWS EC2 NVIDIA A10G DDP
  • Achieved 76.15% Top-1 and 91.87% Top-5 accuracy over 50 epochs
  • Trained using AWS g5dn.12xlarge with 4x NVIDIA A10G GPUs
  • Implemented mixed-precision training for optimal performance
  • Deployed demo on HuggingFace Spaces

Telegram AI Assistant

Reasoning-driven AI Agent

Reasoning-driven Telegram agent using Model Context Protocol (MCP)

MCP Telegram API Google APIs Python
  • Performs web search and document Q&A capabilities
  • Integrated mathematical calculations and reasoning
  • Automated Google Documents and Gmail operations
  • Built using Model Context Protocol for enhanced reasoning

Technical Skills

Programming & Frameworks

Python JavaScript SQL PyTorch FastAPI Streamlit

AI & Machine Learning

Generative AI LLMs Computer Vision NLP Deep Learning MLOps

Databases

PostgreSQL MongoDB MySQL ElasticDB Redis

Cloud & DevOps

AWS Azure Docker Kubernetes CI/CD MLFlow

Let's Connect

Ready to collaborate on exciting AI projects?

I'm always open for exciting opportunities & collaborations! Whether you want to discuss AI innovations, explore project partnerships, or just have a chat about the latest in neural network architectures.

Looking for collaborators to build scalable Advanced Agentic AI Systems!