Projects

PROBLEM

Strava does not show the analytics, comparisons, and visualisations that serious endurance athletes want. No way to compare years, see activity grids, or plan routes with weather overlays.

APPROACH

Built a full analytics platform with Firebase and Next.js, connecting directly to the Strava API. Added AI-powered training insights using the Anthropic API with web search and tool use.

RESULT

Live platform at datavata.app with performance analytics, progress tracking, activity mapping, route planning, and an AI training coach. Used daily. Currently operating under Strava's user limit while awaiting expanded access.

PROBLEM

Ayurvedic health assessments online are either oversimplified quizzes or require expensive consultations. No accessible tool that adapts to individual responses.

APPROACH

Built a conversational AI agent using the Anthropic API that asks structured questions about constitution, digestion, sleep, and temperament, adapting based on previous answers.

RESULT

Live at agniatlas.com. Free to use. Generates detailed dosha profiles with dietary and lifestyle recommendations tailored to each user.

PROBLEM

Music discovery on streaming platforms is passive and algorithmic. No way to connect with people based on genuine music taste rather than demographics.

APPROACH

Built a social music discovery game where users swipe through 30-second previews from mystery playlists and match based on shared taste. Next.js, Firebase, real-time Spotify OAuth.

RESULT

Fully functional product with authentication, previews, matching, and messaging. Paused because Spotify restricts developer mode apps to 5 authorised users.

PROBLEM

Wanted to let people place themselves into street-level photos from anywhere in the world, but the Google Street View API is prohibitively expensive for a free product.

APPROACH

Built with Leaflet maps, Mapillary API for free street-level imagery, and @imgly for in-browser background removal. Custom shadow compositing for realistic results.

RESULT

Working product, but Mapillary has roughly 1% of Google's street-level coverage. Most locations return no imagery, making the experience unreliable.

PROBLEM

Dating apps rely on binary swiping which creates superficial judgements and overwhelming choice. Wanted to test whether ranking profiles would produce more thoughtful engagement.

APPROACH

Built a dating app where users rank 4 profiles from most to least interested instead of swiping. Started with Lovable, moved through Supabase, then Firebase and Next.js.

RESULT

Concept complete with full backend infrastructure. Found that ranking adds friction without improving outcomes. The existing swipe model's speed and simplicity proved superior.