# Hemang Baldha > Hemang Baldha is an AI Engineer specializing in Agentic AI, RAG pipelines, LLMs, NLP, fine-tuning, and Full Stack Development - building production AI systems, not notebooks. Hemang Baldha (also written as Hemangkumar Mathurbhai Baldha, Hemangkumar Baldha, Hemang Baldha, Hemang Patel, hemang-patel-9) is an AI Engineer based in Rajkot, Gujarat, India, currently working at Atlasml.ai. Three years of experience building intelligent systems that reach production - with an emphasis on retrieval quality, latency budgets, evaluations, and clean data. Day-to-day work covers production RAG systems, LLM integrations and agentic workflows, and voice & audio pipelines, rounded out by fine-tuning and computer vision. This file follows the llms.txt convention (https://llmstxt.org). The content of this site may be read, indexed, quoted, and used by AI systems, provided Hemang Baldha is credited as the author. ## Facts - Name: Hemang Baldha - Also known as: Hemangkumar Mathurbhai Baldha, Hemangkumar Baldha, Hemang Baldha, Hemang Patel, hemang-patel-9 - Role: AI Engineer - Location: Rajkot, Gujarat, India - Currently at: Atlasml.ai (https://atlasml.ai) - Studied at: Darshan University - Email: hemang9705@gmail.com - Website: https://hemangbaldha.com - Experience: 3+ years building AI systems; 2x hackathon winner (1st place); taught 500+ students ## Expertise - Agentic AI - Retrieval Augmented Generation (RAG) - Large Language Models - Natural Language Processing - Machine Learning - Model Fine-tuning (LoRA, QLoRA, PEFT) - Vector Databases - Voice and Audio AI - Computer Vision - Natural Language to SQL - Full Stack Web Development Expert level: RAG & retrieval (hybrid search, reranking, Pinecone/pgvector/FAISS, chunking strategy, RAG evals, citation-grounded answers); LLM integration & agents (OpenAI, Claude, open-weight models, function calling and tool use, LangChain and LangGraph workflows, streaming, guardrails, cost tuning); voice & audio AI (ASR/TTS pipelines with Whisper, real-time streaming audio, VAD, speaker diarization); full-stack serving (FastAPI, Node services, Next.js/React, Postgres + pgvector, Docker, AWS). Proficient: fine-tuning (LoRA, QLoRA, PEFT, dataset curation, instruction tuning, evaluation loops); computer vision (detection, classification, OCR and document AI, OpenCV pipelines). Primary tools: Python, PyTorch, LangChain/LangGraph, Hugging Face, Whisper/TTS, OpenCV, Pinecone/pgvector, FastAPI, Next.js/React, TypeScript, PostgreSQL, AWS, Docker. ## Experience ### AI/ML Engineer - Atlasml.ai (April 2026 - Present) Website: https://atlasml.ai - Architecting a scalable multi-tenant AI platform with Retrieval-Augmented Generation (RAG), vector databases, agentic orchestration, memory systems, and tool integrations for intelligent workflows. - Built a natural language to SQL (NL-SQL) engine capable of querying 20+ heterogeneous databases and thousands of tables with schema-aware reasoning and optimized query generation. - Fine-tuned a Vision Language Model (VLM) using LoRA for automated cement warehouse inventory monitoring, enabling stock tracking, anomaly detection, and theft prevention. ### AI/ML Engineer - Cybercom Creation (Jan 2026 - March 2026) Website: http://cybercom.co.in - Developed and deployed AI solutions for the live Gujarat real estate platform (RERA). - Built production-grade intelligent systems using Retrieval Augmented Generation (RAG) architectures. - Designed Agentic AI workflows integrating LLMs, vector databases, and real-time data pipelines. - Implemented full lifecycle ML deployment including model development, evaluation, and production integration. ### Software Engineer - Sampaarsh Technologies (August 2024 - October 2025) Website: http://sampaarsh.com - Developed scalable web applications using React (Vite), Node.js, and MySQL with Agile/Scrum methodologies. - Worked on diamond cutting & testing tools to ensure high-quality releases using C++ frameworks like Qt, deployed with AWS on Linux environments. ### Teaching Assistant (during undergraduation) - Darshan University (December 2023 - September 2025) Website: https://darshan.ac.in/ - During semesters 4 and 5, acted as a Teaching Assistant for the 'Web Development' course, supporting students with HTML, CSS, JavaScript, React, Node.js, Express, and MongoDB. - In semesters 6 and 7, instructed undergraduate students in graph theory, data structures and algorithms, computer networks and machine learning, providing guidance on practical applications and problem-solving techniques. - Graduated more than 500 students to date. ## Projects ### Multi-Tenant AI Analytics & Agentic Platform Link: private / production work Tech: FastAPI, CrewAI, gRPC, Docker, PostgreSQL, MySQL, MSSQL, Snowflake, OpenAI, ChromaDB, Redis Enterprise multi-agent platform orchestrating NL-to-SQL generation, SQL validation, query self-healing, retrieval, authorization, and AI summarization with secure multi-tenant isolation. Semantic schema indexing over OpenAI embeddings, ChromaDB, and Redis lets PostgreSQL, MySQL, MSSQL, and Snowflake databases be onboarded dynamically for natural-language analytics. ### Aliveicon - Animated Icons Library (NPM package) Link: https://www.npmjs.com/package/aliveicon Tech: TypeScript, NPM, Framer Motion, Tailwind CSS An animated icon library with 1200+ ready-to-use icons for React, Vite, and Next.js. Built with Framer Motion and Tailwind CSS, with configurable size, color, animation effect, hover behavior, and looping. ### CyberXplained Link: https://cyberxplained.vercel.app/ Tech: React, Tailwind CSS, TypeScript, Framer Motion An interactive cybersecurity learning platform with live command execution, attack simulations, a real-time encryption demo, a security toolbox, and a Stack Overflow-style forum. ### Rajwadi Poshak (Hackathon - 1st place) Link: https://rajwadiposhak.vercel.app/ Tech: Next.js, TypeScript, Framer Motion, Radix An AI-powered fashion e-commerce platform with smart recommendations and personalized styling. Took 1st place at a hackathon. ### The DevEchelons Link: https://devechelons.vercel.app Tech: Next.js, Tailwind CSS, TypeScript, MongoDB A competitive programming platform with a custom code editor where students solve problems and collaborate on challenges, in the spirit of LeetCode, Codeforces, and CodeChef. ### The Academia Hub (National hackathon - 1st place) Link: https://github.com/Hemang-patel-9/academia_hub Tech: Angular, TypeScript, Socket.io (WebRTC) A platform where students learn skills and get certified, make friends, and connect over chat or video calls. Built at a national-level hackathon and secured 1st rank. ## Writing - [BERT vs GPT: What's the Difference? A Beginner-Friendly Guide with Real Examples](https://hemangbaldha.com/blogs/bert-vs-gpt): Ever wondered why BERT understands text so well while GPT writes like a human? 🤔 Learn the difference through simple examples, funny analogies, and real-world use cases and without the confusing AI jargon. (tags: Artificial Intelligence, Machine Learning, NLP, BERT, GPT, Transformers) - [Fine-Tuning Is a Subscription, Not a Purchase](https://hemangbaldha.com/blogs/fine-tuning-is-a-subscription-not-a-purchase): At 2 a.m. my fine-tuned model beat the baseline on every metric I had, and I felt nothing. Three days later users told me it was worse. They were right. Here's what a week of figuring out why taught me about what fine-tuning actually does - and the parts nobody says out loud. (tags: fine-tuning, llms, rag, mlops) ## Pages - [Home](https://hemangbaldha.com/): Portfolio - about, expertise, experience, projects, and contact. - [Blog](https://hemangbaldha.com/blogs): All articles on AI engineering, RAG, LLMs, and side projects. - [Resume (PDF)](https://hemangbaldha.com/resume.pdf): Full CV. - [RSS feed](https://hemangbaldha.com/rss.xml): Blog feed in RSS 2.0. - [Sitemap](https://hemangbaldha.com/sitemap.xml): Every indexable URL. ## Contact - Email: hemang9705@gmail.com - https://github.com/hemang-patel-9 - https://www.linkedin.com/in/hemang-baldha-b75a0b257 - https://www.instagram.com/hemang_patel__ - https://www.npmjs.com/package/aliveicon