Bhuvan Desai
Applied AI & Data Science @ IIT Jodhpur · SIH 2025 National Finalist
Open to AI/ML & Data Science internships · BengaluruAbout
Second-year B.S. Applied AI & Data Science student at IIT Jodhpur, focused on AI/ML, data science, and backend development. I build real projects end to end — from fine-tuned language models and agentic AI systems to production backends. Currently looking for an AI/ML or Data Science internship in Bengaluru.
Work Experience
Smart India Hackathon 2025
Led the backend for a defence cyber-incident reporting portal (problem ID SIH25183) that reached the national finals of Smart India Hackathon 2025. Built the FastAPI REST APIs for reporting, analysis, and case management — with JWT login, password hashing, and role-based access — and added pipelines to score risk, analyze URLs, and prioritize incidents, backed by PostgreSQL + SQLAlchemy and packaged with Docker.
Raichur Hardwares & Agriwares Pvt. Ltd.
Built software for the family pipe-and-hardware business: a tool that generates pipe price quotes and part of the backend for its online store, raichurhardware.in. Set up daily billing and monthly account checks in Tally Prime and weekly stock counts — finding and fixing recurring data-entry errors — and ran the rollout and training for the tools with non-technical staff.
Latest Blog Posts
Recent articles and insights
Projects
Selected workAn interactive air-quality platform covering 11 Indian cities and 10 Bengaluru stations. A pipeline recalculates CPCB AQI from raw sensor readings and drops faulty values (validated against IQAir), and the analysis is exposed as an 11-tool MCP server (Docker + token auth) with an 'Ask AI' LangGraph agent on Groq's Llama-3.3-70B that answers questions by calling those tools and remembers the conversation.
Fine-tuned Microsoft Phi-3-mini with QLoRA (4-bit, rank 8, alpha 16) to turn plain-English questions into SQL over a sample sales database. Benchmarked the fine-tuned adapter against base Phi-3 to quantify the gain, published the adapter and a live demo on Hugging Face, and tested GGUF quantization for faster CPU inference.
A Streamlit app that scores Telco customers for churn risk and recommends a specific retention action for each one. Compared three models with 5-fold cross-validation, tuned the decision threshold for higher recall, and used SHAP and model coefficients to explain why each prediction was made.
A four-page Power BI report analyzing Blinkit's quick-commerce operations: revenue, delivery SLAs, inventory, and customer retention. Built a star schema model with 45+ DAX measures via Tabular Editor, and caught and documented several data-quality issues in the source data before they could skew the numbers.
Digital Laxman Rekha · SIH 2025
2025Backend for a defence cyber-incident reporting portal that reached the national finals of Smart India Hackathon 2025 (SIH25183). Modular FastAPI REST APIs for reporting, analysis, and case management with JWT auth, password hashing, and role-based access, plus automated pipelines for URL analysis, incident risk scoring, and prioritization — backed by PostgreSQL and packaged with Docker.
Pipe Quotation & Management System
2025A web app that builds price quotes for pipes and hardware: product search, cart-based quote building, GST pricing and configurable discounts, and branded PDF export. Runs in production on the shop's local network (server + client machines), kept always-on with PM2, with role-based logins and admin tracking of every customer's past quotes in MySQL.
A modular, plugin-based AI framework that lets you stitch together models and tools into an assistant that fits you. A LangGraph-orchestrated desktop app (Tauri) with its own plugin and theme registries, so capabilities can be added and themed without touching the core.
A FastAPI web app that generates print-ready showroom sheets with scannable QR codes for products — QR generation with qrcode, image handling via Pillow, and branded PDF export with ReportLab, built to speed up in-store catalogs and labeling.
A simple, no-nonsense uptime monitor that registers URLs, pings them every minute, and tracks their up/down status and response times. Built with FastAPI, asyncio for concurrent pings, and a Docker Compose setup for easy deployment.
A conversational assistant that streamlines bakery order management — taking and confirming customer orders and answering queries to improve overall efficiency and customer satisfaction.
Exploratory data analysis over 186,849 order records, surfacing revenue, timing, and product-bundle insights with Pandas and Matplotlib.
Skills
Education
Indian Institute of Technology, Jodhpur
B.S. in Applied AI & Data Science (Hybrid)
CGPA 8.0 / 10 (through Semester 4). Coursework: Machine Learning, Pattern Recognition, Linear Algebra, Probability & Statistics, Numerical Optimization, Data Visualization, Databases, Algorithmic Thinking.
Vedantha PU College, Bengaluru
Pre-University — Science (PCM + Computer Science)
Scored 82.38% in the Karnataka PUC board examinations.
Certificates
Completed courses and professional certifications
AI Engineer: Agentic AI, MCP & Multi-Agent Systems
Built multiple autonomous agents — a personal knowledge chatbot, an SDR agent for automated sales outreach, and a deep-research multi-agent system — using LangGraph, MCP, and tool-using workflows for decision support. Issued by Udemy (Ed Donner).
AI Engineer: LLM Engineering, RAG & Agents
Developed diverse LLM applications including intelligent web-scraping agents, multimodal customer-support systems, and audio-to-insights pipelines, and evaluated trade-offs between RAG, fine-tuning, and agentic workflows across frontier and open-source models. Issued by Udemy (Ed Donner).
Advanced Large Language Model Agents
UC Berkeley MOOC on LLM-based autonomous agents — reasoning, planning, tool use, inference-time reasoning, function calling, and code generation.