Sre Varshan

Showcase

Visual Archive

A 3D gallery of my engineering journey. This sphere holds snapshots of the competitions I've tackled, the places my work has taken me, and the physical builds I am most proud of. Drag to rotate and click to view

Meet Sre Varshan

SRE
VARSHAN

Gen AI Engineer.

I apply AI across multidisciplinary fields to boost efficiency, building practical solutions that real people are actually using at scale.

BASED IN

13.0827° N, 80.2707° E

Tech Stack

SKILL SET & CAPABILITIES

Sketch of Sre Varshan at a laptop surrounded by his tools: Hugging Face, PyTorch, AWS, Google Cloud, Docker, C, NVIDIA, Amplify, OpenRouter, Claude, SQL, REST API, GitHub, LangChain, Ollama, Git, NumPy, pandas, PostgreSQL, Python, scikit-learn and vLLM

Work experience · 02 roles

Real work, real impact.

Every role I've taken has been about building something that genuinely works for real people — not just demos.

Healthcare SaaS · IoT + machine learning · Remote

01 / EmedLogix

Intelligent System Developer

Worked as an AI Product Developer Intern at EmedLogix (US-based healthcare technology SaaS), contributing to a production provider onboarding and credential management system active in the market.

What I built

Gen AI inside a real credentialing platform

Healthcare credentialing is slow, manual and document-heavy. I worked on bringing Gen AI into a production credentialing application end to end, handling sensitive, real US healthcare provider data under strict monitoring.

Text-to-SQL Agent
My core role. I developed an agent that converts plain-English questions into SQL and retrieves the data. The agent knows the database schema of the PostgreSQL database on AWS RDS.
OCR Pipeline with Qwen2.5-VL 7B
Built an OCR pipeline on Qwen2.5-VL 7B for accurate text extraction across varied documents, autofilling form fields and saving a lot of time compared with traditional manual filling.
Document Deadline Tracking
Every provider carries 60+ documents. I added automated deadlines for each document of every provider to track its lifecycle and renewal cycle, with alerts 30 days before it falls due.
CAQH & NPI APIs
Worked with the real CAQH and NPI APIs from the US to retrieve and validate provider information.
Secure Sign-in
Integrated Azure Entra ID SSO login through a web service and added Cloudflare bot verification for additional security.
Login & Landing Page Design
Worked on the frontend design of the login page and landing page to make them user friendly.
Beta Deployment on AWS
Deployed the beta version of the application to the company on AWS during the internship.

Key highlights

Sensitive US healthcare provider data, under strict monitoring
Real CAQH and NPI API integrations
Beta version deployed on AWS during the internship
Azure Entra ID SSO with Cloudflare bot verification

Stack

Generative AI
Text-to-SQL Agent
Qwen2.5-VL 7B OCR
CAQH & NPI APIs
FastAPI / Python
PostgreSQL on AWS RDS
Microsoft Azure Entra ID SSO
Cloudflare Bot Verification
AWS Beanstalk / RDS / Amplify
Frontend: Login & Landing Pages
SaaS Production Architecture
Team-based Agile Development

02 / Microsoft × Edunet

AI/ML Intern

A virtual internship with Microsoft and the Edunet Foundation: core machine learning on the Microsoft Learn portal, from the basics to intermediate algorithms and techniques, finished with a capstone project of my own.

What I built

AI Nose, the capstone

Microsoft Learn gave me quality courses on core machine learning, from the basics to intermediate, and introduced me to a range of ML algorithms and techniques. One algorithm caught my interest: the Random Forest classifier. With my background in IoT and hardware, the idea for the capstone followed: an AI nose that can actually smell its environment, detect particular gases, classify them in real time and alert when a gas exceeds a safety threshold.

IoT Hardware & Sensors
Programmed an ESP32 microcontroller integrated with MQ3 and MQ9 gas sensors. These sensors measure concentrations of alcohol vapors, carbon monoxide, methane, LPG, and hydrogen.
Signal Processing & Cloud Sync
Developed initial ESP32 preprocessing routines for sensor calibration, value normalization, and noise filtering. Timestamped datasets were streamed in real time to Firebase Realtime Database for instant synchronization.
Odor Level Classification Model
Collected a labeled sensor dataset categorized into Clean, Moderate, Foul, and Very Foul. Tested various machine learning models to choose the most reliable and fastest one, settling on a decision-tree-based algorithm (Random Forest) that achieves 98% accuracy and runs in milliseconds on the server.
Live Monitoring Web Dashboard
Built a Gradio web application visualizing live sensor readings, predicted odor classes, and historical trends for remote monitoring.
Automated Sanitation Alerts
Programmed alerts that trigger whenever predicted levels are Foul or Very Foul, notifying maintenance teams immediately so they can clean proactively rather than on a fixed schedule.

Key highlights

Hardware: ESP32 Microcontroller
Sensors: MQ3 & MQ9 Gas Sensors
Database: Firebase Realtime Database
Dashboard: Python Gradio Visualizer

Stack

Microsoft Learn: Core ML
Random Forest Classifier
Embedded ESP32
IoT Sensor Integration
Firebase Cloud Services
Gradio Dashboards
Python Data Science
Feature Engineering
Odor & Gas Calibration
Predictive Maintenance

What I've Built

Not stopping in the classroom — taking concepts from theory and engineering them into real-world deployed solutions.

PATENTED AI

Banana Weevil Detection project using Deep Learning

A dual-mode acoustic sensing system that detects Odoiporous longicollis infestations in banana crops entirely offline.

Offline Inference
Acoustic dual-sensing
Low-power Raspberry Pi
Government Patented

1ST PRIZE · SPECTRUM '25

AgroCare — Multimodal AI & IoT for Cotton Farming

An award-winning precision farming platform combining a custom field probe, drone crop imagery, CNN disease detection, and LLM-guided recommendations.

Custom pH, NPK & moisture probe
ESP32 + LoRa field telemetry
Drone crop imaging + SLAM
CNN disease detection + LLM advice

LOCAL OCR FRAMEWORK

TextLens — GPU-Accelerated VLM OCR Framework

A Python framework that makes local document OCR simple to set up, reuse, and run across images and PDFs.

Local image & PDF OCR
GPU-aware model setup
Batch processing & caching
FastAPI OCR serving

AI NUTRITION PLATFORM

NutriMinds AI

An AI-powered nutrition intelligence platform analyzing medical reports and food labels via Vision-Language models and fine-tuned LLMs for personalized dietary analysis.

VLM OCR & Reasoning
Fine-tuned Mistral 7B
Open Food Facts Data
Clinical Nutrition Rules

RAG FROM SCRATCH

RAG Pipeline From Scratch

A fundamental document-grounded question-answering pipeline built hands-on to understand retrieval, embeddings, and context generation.

Multi-format document ingestion
Chunking & preprocessing
ChromaDB similarity retrieval
Groq-grounded answers

PREDICTIVE GOVERNANCE

ACAS DHRISTI

A predictive governance framework for UIDAI proactively forecasting stress at 173,225+ Tamil Nadu Aadhaar centers 7 days in advance.

173,225 centers mapped
7-Day Stress Forecast
n8n autonomous routing
CIDR Privacy Compliant

AI-POWERED CRM

Xenia CRM

An enterprise retail CRM with automated AI voice outreach (ElevenLabs), Llama 3.3 campaign generation, and Gradient Boosting churn prediction.

AI Voice Campaigns
Llama 3.3 Copywriting
GBM Churn Classifier
Event Loop Simulator

Coding & Social Profiles

Where the work lives

Code, write-ups and demos

Where I sharpen it

Competitive programming & practice

Download Resume

Highlights

The Trophy Room

A few things I'm proud of — from government patents to hackathon wins and community impact.

1st Place — Spectrum '25

VIT Chennai | $500 Prize

Won first place at the Spectrum '25 National Hackathon for our innovative AI solution, competing against top university teams.

1st Place — SolveX AI Challenge

SolveX AI | Rs. 5,000 Prize

Awarded first place for building outstanding AI prototypes addressing real-world operational challenges.

2nd Prize — Video Making

NIFTEM Tanjore | Rs. 3,000 Prize

Won second prize in the national video production competition for creative storytelling and execution.

Top 8 National — NAF Uzhavu

IIT Madras | IIM Mentorship

Recognized as a Top 8 national finalist at IIT Madras and selected for the prestigious IIM Product Commercialisation Mentorship.

IIT Goa National Workshop

5-Day AIML & Drone Training

Completed an intensive hands-on national level workshop at IIT Goa focused on Autonomous Drones and Edge AI/ML applications.

IIC Regional Meet 2025

Best Innovation Across South

Awarded the Best Innovation accolade across the Southern region at the Thirunelveli regional meetup.

Dindugal Science Fest

7-Day Pitching & Networking

Successfully pitched our edge computing product to CEOs, founders, high-ranking government officers, and domain experts.

AI School Outreach Program

Government School Careers in AI

Organized educational sessions for higher secondary students to introduce AI careers and provide hands-on experience with edge hardware.

Government Patent Published

App No. 202541116228 A

Filed and published a government patent for a novel acoustic-sensing system for early banana weevil detection on edge hardware.

Organized TENSOR'26 Hackathon

National Level AI Hackathon

Core organizer and developer managing infrastructure, scheduling, and participant experience for a national-scale hackathon.

FOSS Design Head

SRM IST Trichy FOSS Society

Led the design team for the Free and Open Source Software (FOSS) society, creating branding and UI assets for community projects.