Headline

AI Engineer | LLM & RAG Systems in Production | Published Researcher | Building AI for Indian SMBs

About

AI Engineer building production LLM systems, RAG pipelines, and document AI. Two published papers in deep learning. Most of what I've built is running in real businesses today. What's in production ConstructionMartt: a multi-role ERP for a construction materials business. Orders, partial dispatch, approval workflows, payments, lead follow-ups with push notifications, real-time sync across sessions. Four user roles, installable PWA, six months live with 100+ customers and 20+ orders a day. ShopMind: a POS system for retail with OCR-based invoice entry and natural-language querying in Hindi and English (text-to-SQL over the shop's data). Deployed and in daily use. Invoice parsing pipeline for a saree retailer using the Claude SDK, turning vendor PDFs into structured purchase entries. Earlier work At Sambhav AI & Tech Services I built a Whisper + OCR video processing pipeline on Docker/Kubernetes that cut processing time by 50%, integrated multimodal AI into an EdTech platform, and reworked the AI search layer for a 75% speed improvement. Research Skin Cancer Detection using Deep Learning, IJCNIS Vol 16 Issue 5 (2024) Optimising Job Search Using Web Scraping, IRJET Vol 11 Issue 4 (2024) Stack LLMs and RAG: Claude, GPT-4o, Gemini, LangChain, vector databases, function calling ML and CV: PyTorch, CNNs, Transformers, OpenCV, Whisper, OCR Backend: Python, FastAPI, MongoDB, WebSockets Deployment: Docker, Kubernetes, Vercel, Render 1st Runner-Up, Hackengers Hackathon (Livewire, Chennai). Former Technical Head, Coding Club. Open to AI Engineer roles. Happy to talk.

Location

Nagpur

AI stack

anthropic apis

claude code

claude

cursor

elevenlabs

gemini

Work experience

Independent AI Engineer at Self-Employed, Sep 2025 - Present · 1 yr 3 mos, Nagpur, Maharashtra, India, Building and deploying production AI systems for small and mid-sized businesses in India. ConstructionMartt: multi-role ERP for a construction materials trader. Orders, partial dispatch with short-close, multi-level approval workflows, customer-wise payments with instalments, lead management with automated follow-up reminders and push notifications, WebSocket real-time sync. Four user roles, installable PWA, FastAPI + MongoDB. Six months in production, 100+ customers, 20+ orders a day. ShopMind: retail POS with OCR-based invoice entry and natural-language querying in Hindi and English (text-to-SQL over live shop data). Deployed and in daily use. Invoice parsing pipeline for a textile retailer using the Claude SDK, converting vendor PDFs into structured purchase entries.

AI Engineer Intern at SAMBHAV AI & TECH SERVICES, Feb 2025 - Aug 2025 · 7 mos, Nagpur, Built and shipped production AI systems across video processing, search, and multimodal integration. - Designed a Whisper + OCR video processing pipeline deployed via Docker and Kubernetes, reducing processing time by 50% - Integrated multimodal AI capabilities (LLM + vision + audio) into an EdTech platform for content understanding and generation - Optimized AI search infrastructure, improving query response speed by 75% - Worked across the full ML stack: model selection, inference optimization, containerization, and production monitoring Stack: Python, PyTorch, Whisper, OCR (Tesseract, EasyOCR), Docker, Kubernetes, FastAPI, REST APIs

Education

Bachelor of Technology - BTech, Computer Science and Engineering, Karmaveer Dadasaheb Kannamwar Engineering College

Public links

github: https://github.com/harshaldonarkar

medium: https://medium.com/@harshaldonarkar

harshaldonarkar.github.io: https://harshaldonarkar.github.io/