Dhruv Lohar · Developer · Pune, India
Third-year CS student at VIT Pune and full-stack and distributed-systems engineer. At JBI, I build a production CRM spanning 5 branches and 5,000+ leads. Previously an SWE intern at NextDR.ai. Runner-up at Flipkart Gridlock 2.0 and rank 29 among 32,000+ teams at Amazon ML Challenge 2026.
This one actually works. Type help and poke around.
I'm a third-year CS student at VIT Pune (2024 to 2028), specializing in IoT, Cybersecurity and Blockchain. I build full-stack applications and distributed systems, from production CRM workflows to real-time monitoring and multi-LLM tools.
At JBI, I work on a CRM for Sunrays Engineering and Solar Tech across 5 branches, designed for up to 500 users. Previously, I spent 11 months at NextDR.ai building disaster-recovery tooling. Outside work, I compete in programming and ML challenges, publish research, and develop patent ideas.
Hover a skill to see where I have used it. No inflated bars.
C++ · TypeScript · JavaScript · Python
React · Next.js · React Native · Tailwind CSS · MapLibre
Node.js · Express · PostgreSQL · MongoDB · Prisma · Socket.IO
Linux · Docker · Google Cloud Platform · Git · GitHub Actions · CI/CD · Vercel
RAG · Embedding-Based Retrieval · Vector Search · Semantic Search · LLM Orchestration · Prompt Engineering · Model Evaluation
JWT · OAuth 2.0 · NextAuth.js · REST API Security · Authentication · Authorization
Things Dhruv Lohar has built and actually uses.
My team took second place out of 33,600 participants. PRAHAR is a traffic intelligence platform for Bengaluru: it forecasts congestion severity, junction risk and how long incidents take to clear, then recommends where to send officers, barricades and diversions. Built on real traffic-police data, pitched live at Flipkart HQ.
A learning tool that renders topic relationships as you type. The streaming knowledge graph uses Canvas 2D, a physics simulation and an LRU cache to handle 100+ nodes at 60 FPS. A shared streaming layer supports Gemini, OpenAI, Claude, Groq and Ollama, with per-provider fallback.
Running a big model on every transaction is expensive, so this routes payments through confidence-gated tiers: cheap calibrated LightGBM first, heavier PyTorch models only when the score is uncertain. Trading off inference cost against recall was the interesting part; calibration curves don't lie.
Combines AST normalization, token comparison, and embedding search to surface similar files and functions. Drop in a repo or paste code, get an exportable report. Built it after wondering how much of my old coursework I'd reused from Stack Overflow.
Built in 24 hours at Crescendo, VIT Pune's software hackathon, where it took first place out of 400+ teams. RollCall layers QR codes, geolocation, Bluetooth chaining between nearby verified devices, and face recognition to address attendance fraud. React Native up front, Express with Postgres and Prisma behind it.
Emergency help button, a live safety map with reported incidents and safe zones, plus a small mindful-journal corner. React Native on Expo with a Dockerized backend. The map work (incident popups, zone coordinates round-tripping to the API) taught me more than any tutorial.
Built a production CRM for Sunrays Engineering and Solar Tech across 5 branches, designed for up to 500 users. The lead pipeline covers 5,000+ leads with assignment history, follow-up tracking and revenue attribution. Added PostgreSQL dashboards, branch-level RBAC, lead ownership, restricted site-engineer access, Google Sheets sync via webhooks and background workers, and Docker deployment with CI/CD.
Ranked 29 among 32,000+ teams in Amazon ML Challenge 2026, earning an Applied Scientist Intern interview opportunity.
2nd of 33,600 participants with PRAHAR, a traffic intelligence platform built on real Bengaluru Traffic Police data. It forecasts congestion severity and incident resolution time, then recommends where to send officers, barricades and diversions. Pitched live at Flipkart HQ.
Disaster-recovery tooling. Replaced polling with SSE monitoring (status latency from 2s to under 100ms), built multi-step recovery orchestration with retries, backup scheduling with retention cleanup, Twilio SMS and call alerts, PDF audit reports, and a JSON-tree UI for infra config.
First place out of 400+ teams with RollCall, built in the 24-hour final. Shipped a React Native app and Express API with Prisma/Postgres, combining QR codes, geolocation, Bluetooth proximity checks and on-device face matching.
Won first place competing against teams from across Pune. Fastest I've ever gone from idea to working demo.
IEEE publication at ESCI 2026, Pune: Multi-Layered Hybrid Framework for Source Code Plagiarism Detection Against AI-Enhanced Obfuscation. Two additional papers under review. Two patents filed: an IoT-enabled floating buoy for real-time microplastics detection in water, and a privacy-preserving face authentication system.
300+ problems solved on LeetCode, with a rating of 1,631. Maximum Codeforces rating: 1,252. DSA keeps the fundamentals sharp for everything else.
Contact Dhruv Lohar by email. Fastest way to reach me is email. I usually reply within a day.