I'm Shorya Sharma — a GenAI Developer Specialist focused on Agentic AI, Multi-Agent Systems, MCP (Model Context Protocol), and Retrieval-Augmented Generation. I design and ship enterprise-scale GenAI applications across Azure & AWS — from RAG pipelines and fine-tuned LLMs to autonomous agents — using Python, FastAPI, Docker, Azure OpenAI, and Azure AI Search.
From multi-agent orchestration to LLM fine-tuning and AWS deployments — I build AI products that actually ship to production.
I'm a GenAI Developer Specialist at Hexaware Technologies (Pune), where I design and build enterprise-grade Generative AI applications, LLM-powered systems, and agentic AI workflows. My core focus is on Agentic AI architectures, Model Context Protocol (MCP) implementations, and multi-agent orchestration that solve real enterprise problems at scale.
Previously at Mantra Smart Identity (Mantra Softech), I built next-gen agentic AI applications for identity & biometric solutions. Before that at L&T Technology Services, I designed and deployed GenAI & Computer Vision models for surveillance, law enforcement, and enterprise NLP — taking ideas from POC to fully operationalized production. I work across the full stack: data pipelines, LLM fine-tuning (LoRA / QLoRA), RAG with vector DBs, FastAPI services, Docker, and dual-cloud deployment on Azure (Azure OpenAI, Azure AI Search, Azure ML, Azure Agent SDK) and AWS (Lambda, SageMaker, ECR).
With 7+ years across IoT, manufacturing, and AI engineering, I bring an engineer's discipline to AI: production-grade APIs, observability, CI/CD, and clean architecture.
Tools and frameworks I use to design, build, and ship production AI systems.
7+ years of engineering — from IoT and manufacturing to leading AI initiatives.
Hexaware Technologies · Pune
Mantra Smart Identity Pvt. Ltd. (Mantra Softech) · Ahmedabad
L&T Technology Services · Ahmedabad
TwiLearn Edtech Pvt. Ltd. · Remote
A curated selection of production-grade AI work across Agentic AI, MCP, GenAI, RAG, MLOps, and Computer Vision.
Production-grade multi-agent AI platform for NSE (India). Specialized agents collaborate to analyze stocks, evaluate fundamentals, technicals, and sentiment, and produce institutional-grade reports.
Autonomous semantic reporting platform: converts natural language to SQL, executes against your DB, and renders interactive ECharts dashboards. Built on FastAPI + LangGraph + React.
RAG-based chatbot built with LLMs, LangChain, and FAISS / Weaviate to intelligently analyze and answer queries from multiple RFP documents. Enables semantic search, scope & compliance extraction, RFP comparison, and bid / no-bid decision support — reducing manual review time by ~70%.
AI-powered employee assessment platform using LLMs and RAG to dynamically generate questions, evaluate responses, and provide context-aware scoring & feedback. Built scalable Python & FastAPI backend services for real-time assessments aligned with hiring and upskilling needs.
Custom MCP server implementations exposing tools to LLM agents — including a Data Preprocessing Agentic System where an LLM agent analyzes uploaded CSVs and a Python execution agent (via MCP) runs cleaning & transformations.
Full-stack autonomous agent system integrated with a FastAPI backend, containerized with Docker for seamless deployment. Modern microservice design + LLM capabilities to automate complex workflows.
End-to-end Retrieval-Augmented Generation pipeline using LangChain, HuggingFace embeddings, and a Q&A chatbot built over the "Attention Is All You Need" research paper for grounded, citation-backed responses.
PEFT QLoRA fine-tuning of Mistral 7B on a Text-to-SQL dataset. Produces a domain-specialized GenAI model that converts natural language queries into accurate SQL.
Agentic AI-powered Streamlit chatbot using a React Agent + RAG, integrating FAISS and DuckDuckGo Search for dynamic retrieval. Uses LangChain, HuggingFace embeddings, and ConversationBufferWindowMemory. Deployed on Hugging Face Spaces for scalable, real-time conversational access.
Experiments combining classical ML pipelines with LLM-powered agents that select features, tune models, and explain results — moving traditional ML towards autonomous, agent-driven workflows.
Reproducible SOP for deploying a Python 3.12 AWS Lambda using Docker, the Serverless Framework, ECR, and API Gateway. Solves manifest incompatibility for Linux/AMD64 Lambda Docker images.
Streamlit chatbot powered by Google Gemini API, containerized via Docker and hosted inside an Ubuntu VM (VirtualBox). Demonstrates full-stack VM deployment of a production LLM application.
Autonomous web-browsing agent that controls a real browser to navigate sites, extract information, and complete multi-step web tasks — bringing tool-augmented agents to real-world browser automation.
Reference architecture for production ML pipelines: data ingestion, validation, training, evaluation, registry, deployment, and monitoring — building a foundation that scales from notebook to production.
GenAI application using open-source LLMs to summarize long product reviews and analyze sentiment at scale — surfacing actionable insights for product quality improvement.
Collection of CV projects: facial expression recognition (FER-2013) with CNNs & OpenCV, face detection, and image-processing pipelines covering happiness, sadness, neutral, and anger detection.
Curated ML portfolio including Human Activity Recognition from smartphone sensor data with 563 features, smart feature selection, and end-to-end classification pipelines on real datasets.
Continuously learning to stay at the frontier of AI engineering.
Open to GenAI / Senior AI Engineering connections, consulting, and collaborations on Agentic AI, MCP, and Generative AI products.