Research first, cloud second, then everything I've shipped and kept running. In priority order.
Research
Semi-Supervised Oral Cancer Image Classification Using GAN-Based Augmentation and EfficientNet-B3
2025 3rd DMIHER International Conference on AI in Healthcare (IDICAIHEI), IEEE Xplore, November 2025.
I led the research, built the machine learning side, and presented the paper to the conference judges at DMIHER 2025. An EfficientNet-B3 classifier trained with a K+1 SSGAN for binary classification under low-annotation settings.
93.5% accuracy, AUC 0.95. A ProtoNet few-shot setup reached 94% on 5-way 10-shot tasks.
I build and ship cloud projects with the group, and served as an evaluator at Deploy or Die 2026, judging teams on Specification & Architecture against a formal rubric.
Zap
February 2026
Serverless peer-to-peer and cloud file transfer
Built with GDG Cloud KIIT. Fully serverless on Cloudflare Workers with R2, running at the edge with no origin server. Direct browser-to-browser transfer over a WebRTC DataChannel, with the Worker used only for signalling.
A three-person team with Sriyam Dasgupta and Ankit Kumar. Talking to doctors at nearby hospitals surfaced the problem: patient data sits siloed inside individual departments, causing fragmented care and repeated diagnostics.
Rather than build an EMR from scratch, we deployed OpenEMR on a Google Cloud VM and made it reachable for remote stakeholders. The hard part was API integration. Jerry Padgett of the OpenEMR community mentored us through it, and we got full API integration working through Postman, plus a custom API module on oe-module-api-explorer with OAuth-secured auth and role-based access. Phase 1 shipped.
Production Postfix/Dovecot on Debian: Let's Encrypt TLS, Fail2Ban, full IPv6, and correct MX, SPF, DKIM and DMARC records.
Debian, Postfix, Dovecot, Certbot
Applied ML & tools
KUSH
March 2026
Bronze award, INDCON
IIE Chennai Chapter / CEG, Anna University, national level. An AI system that predicts supply chain disruptions before they happen. I worked the implementation side, building the ML models and the logic behind the system. With Utkarsh, Shoham Chakraborty and Kavish Basu.
statmux
2026 to present
Developer analytics dashboard
Unifies GitHub, Codeforces and LeetCode into one profile with a computed Code Health score and a public page at /u/[username]. Three inconsistent third-party APIs normalised behind one Express service, with cached refreshes to stay inside upstream rate limits. Full SDLC documentation: SRS, UML, test cases, RTM.
An Electron desktop app plus browser extension for media discovery and playback. Spawns MPV and VLC directly with automatic path resolution and hardware acceleration, includes a multi-stream downloader with live transfer metrics, a headless-Chromium resolver for intermediate redirect gateways, remote configuration over the Supabase REST API, and .m3u playlist export. Runs on Linux, Windows and macOS.
160+ GitHub stars, packaged for the Arch User Repository and maintained across releases. Reverse-engineered an undocumented server-rendered JSON feed, using BeautifulSoup to find the embedded payload and regex to pull it out. That drives a detrended sparkline of engagement hotspots.
Python, Shell, BeautifulSoup
XYlofy AI
Remote, April to May 2026
Data analyst intern
End-to-end sentiment analysis pipeline over the Amazon Fine Food Reviews dataset (500K+ reviews): cleaning, TextBlob polarity scoring and visualisation, delivered as a reproducible Jupyter notebook.