Founder of Flickstat·Final-year CS student·Open to opportunities
// 01 — Stack
// 02 — Work
Football analytics platform providing live match data and predictions. Founded and led a 3-person team, resolving canonical IDs across 4 data sources.
✦ 1,000+ pages served via SSR/SSG · Real users · Live WebSocket data
// origin
Azeem originally built this alone under the name "Lens." He scraped everything, built the pipeline, but hit a wall at canonical ID resolution — when Rashford moved from Manchester United to FC Barcelona, he had two completely separate IDs across data sources. None of the sources agreed on player or club names either (Man United vs Manchester United, accented names, aliases). He deleted everything, restarted from scratch, and brought in two college friends — Arnav and Vedant — explaining the entire vision to them and co-designing the schema from zero.
// under the hood
Data is combined from five sources — WhoScored, Understat, BSD, Transfermarkt, and TheSportsDB (images). Each source uses its own internal IDs. The team designed a canonical "Flickstat ID" system that maps every player and club across all five sources into one unified identity. Automated daily scraping runs via a run_daily_pipeline triggered by cron-job.org firing a GitHub Actions workflow. Live match data and odds are delivered via WebSockets through BSD running on an Oracle Cloud VM. The site serves 1,000+ pages via SSR and SSG — one of the real production challenges was deciding which rendering strategy to use where, and getting that wrong initially. Azeem owns the backend; Vedant handles frontend; Arnav manages automation.
Fully offline Windows desktop AI app for wildlife conservation, processing 3,000+ files locally without internet.
✦ Fully offline · MegaDetector v5a · 2,400+ species identified
// origin
Azeem's cousin is a research student doing field work — planting GoPros and camera traps at wildlife sites in Gujarat, then manually labelling hundreds of videos afterward. One conversation sparked the idea: what if the labelling was automated, entirely offline, on the laptop they already carry? The target user is student researchers and field conservation teams, particularly those working with Indian species.
// under the hood
The hardest part wasn't the ML — it was finding the right combination of models. After testing several options, the final pipeline uses MegaDetector v5a for animal detection and flagging, Google SpeciesNet for species identification, and an OpenCLIP model that gave better results specifically for Indian species in Gujarat. Packaging all of this into a single offline .exe installer (PyInstaller + Tauri v2) without internet dependency was genuinely difficult. One feature does optionally use the Gemini API — reading timestamps from video frames — but the core pipeline (detection, species ID, flagging, CSV export) works completely offline. The output is a clean CSV with every frame labelled and flagged, ready for research use. Currently tested end-to-end by Azeem; first external handoff to his cousin is upcoming.
Browser-only bulk image processing using Canvas API and Web Workers.
// origin
Built for creators and companies who process images in bulk — designers, social media teams, small businesses. The core design decision was zero server uploads: everything happens natively in the browser. This wasn't just a performance choice — it was a deliberate privacy stance. People are rightfully skeptical of uploading images to unknown servers (think: PDF editors selling your data). On Batchify, nothing leaves your machine.
// under the hood
Processes up to 150 images simultaneously in-browser using the Canvas API and Web Workers. The tricky engineering problem was mobile: mobile browsers cannot handle compute-heavy operations like smart crop or background removal via Web Workers — they simply don't have the resources. The solution was device detection to gracefully degrade on mobile while keeping the full feature set intact on desktop. Fast, private, and honest about its constraints.
Full-stack Progressive Web App (PWA) expense manager built for modern teams.
// 03 — About
"I couldn't finish a single project. Then one day, I did."
In the beginning, I couldn't finish a single project. I'd start with grand ideas, get stuck in the weeds, and eventually abandon them. That was my reality for a long time. But over time, I learned to push through the friction. Today, I build and maintain multiple live projects with real users.
The turning point was Flickstat. It was the hardest thing I've ever built. I originally tried to build it alone, but quickly realized the scale required a team—so I brought in two friends and led us to launch. The deepest technical challenge wasn't just shipping; it was canonical ID resolution. We had to map players and clubs across four completely different football data sources into a single, unified identity. It was messy, grueling work—and we solved it. We now serve thousands of pages via SSR/SSG, deliver live match data and odds through WebSockets, and run a real-time prediction leaderboard.
I also care deeply about software that matters in the real world. That led me to build WildWatch, a fully offline AI desktop app for wildlife conservation teams. No cloud, no API keys—just detection models running locally on field laptops.
I am currently a final-year Computer and Communication Engineering student at Manipal University Jaipur (CGPA 8.68). Based in Rajasthan, India, I'm actively looking for opportunities to tackle hard, meaningful engineering problems.