Harsh S- Senior GenAI Engineer

Major Projects

AI-Powered Symptom Checker Bot – Intelligent Triage for Respiratory Diagnosis

Developed an AI-powered Symptom Checker Bot to triage respiratory conditions through human-like conversations. Built a hybrid rule-based and dynamic questioning workflow using FastAPI to drive accurate symptom interpretation and diagnosis. Integrated Azure OpenAI (ChatGPT API) for natural language understanding and emergency detection. Deployed the system on AWS EC2 with Docker and GitLab CI/CD for scalability and uptime. Collaborated with domain experts to refine diagnosis logic and improve triage precision.

USA

Frontend:

React.js

Backend:

FastAPI
LangGraph

AI:

GPT- 4o
Rule-based logic

Infrastructure:

AWS EC2
Azure OpenAI API
CI/CD
Docker

AI Chatbot for Tyre Recycling Operations – Automating Customer Engagement & Workflow Management

Designed and delivered a domain-specific AI chatbot to enhance operational efficiency and customer interaction for a tyre recycling enterprise. Developed a multi-agent architecture using LangChain and integrated LLaMA 3-70B via GROQ to handle natural language queries with high accuracy. Built intelligent modules for appointment scheduling and management (create, update, delete, retrieve), automated invoice generation, and tyre price estimation from uploaded images. Orchestrated structured conversational workflows and NLP pipelines for end-to-end automation.

USA

Backend:

Python

AI:

LangChain
LLaMA 3-70B
Vector DB

Infrastructure:

GROQ

Generalized Reinforcement Learning Framework – Dynamic Environment Configuration for Scalable Experimentation

Enabled dynamic reinforcement learning environment creation through a JSON-configurable framework that supports rapid experimentation across diverse industrial use cases. Architected a modular backend to parse and deploy custom simulation scenarios, eliminating rigid, hardcoded setups. Integrated multiple RL algorithms with environment compatibility, emphasizing scalability, adaptability, and reproducibility for evolving workflows.

USA

Framework:

Gym
SB3

Legal Documents Summarization System – Accelerating Legal Review with Fine-Tuned NLP

Streamlined legal document processing by implementing a transformer-based summarization system tailored to the legal domain. Utilized BERT fine-tuning to generate concise summaries that preserved legal accuracy, reducing document length by 70% without losing critical meaning. The system significantly accelerated review cycles, improving decision-making speed by 50% and cutting overall review time by 40%, resulting in a 25% boost in team productivity. Legal experts validated summarization outputs to ensure compliance and contextual relevance.

Spain