Hyperfocused
Engineer

On-device AI for high-stakes conversations — your AI that knows who you're talking to before you do.

CORTIS is a context intelligence platform that ingests your life context — calendar, email, contacts, LinkedIn, CRM — and delivers actionable intelligence before and during every conversation. Not meeting notes. Not post-call analysis. Live runtime intelligence delivered through phone, earbuds, or smart glasses. CORTIS targets the gap between what you know and what you can recall in the moment.

Compose a post once, customize per platform, and dispatch to Twitter, LinkedIn, Reddit, Instagram, WhatsApp, and Facebook at once — with a vintage writing-desk aesthetic. → Details

I needed to step up my posting across platforms while building in public, but reformatting and manually posting the same content to six different platforms was killing momentum. Rather than paying for every API and AI provider upfront, I designed a pragmatic tiered system — OAuth where available, native share sheet as fallback, and a chatbot paste-back workflow for AI reformatting that costs nothing. The result is an app I use daily with a vintage writing-desk aesthetic that makes every post feel intentional.

Master system design concepts through interactive, arcade-style mobile games. Learn caching, sharding, load balancing, and more — by playing. → Details

System design interviews are notoriously hard to prepare for — most resources are walls of text or hour-long videos. I wanted to learn by doing, so I built an arcade-style mobile game that teaches concepts like caching, sharding, load balancing, and database design through interactive challenges. Play as a guest with zero data collection, or sign in to sync progress across devices.

Run professional-grade cricket auctions with real-time bidding, dual auction models, role-based live views, club management, and tournament hosting. → Details

Cricket auctions — especially IPL-style ones — require coordinated real-time bidding, budget tracking, and instant player assignment. Running them on spreadsheets is chaos. I built a full auction platform with two bidding models (sealed-tender and open outcry), role-based live views for auctioneers, captains, and spectators, and real-time Supabase subscriptions powering every update. It grew into a complete cricket management platform with club creation, event scheduling, tournament hosting, and a public discovery gateway.

Track feeding, sleep, diapers, and milestones with voice logging, WHO growth percentiles, and collaborative caregiving through real-time family sync. → Details

Parenting data is scattered — feeding logs in one app, sleep schedules in another, photos on phones, no way for grandparents and caregivers to stay in sync. I built Koo as a unified hub: log activities by voice or form, track growth against WHO percentile curves, share milestones with family through a private Village feed, and give each caregiver role-based access. Offline-first with real-time Supabase sync when connected.

A cyberpunk-themed mobile app that teaches Python, PyTorch, and AI/ML through gamified coding missions with an AI co-founder.

Zeroshot turns coding education into a cyberpunk narrative. Instead of passive tutorials, you complete missions — writing real code guided by Nova, your AI co-founder. Starting with Python fundamentals, the journey progresses through data structures, PyTorch, and into LLMs and AI/ML concepts. Each sector has Easy, Medium, and Hard difficulty tiers with star-based progression.

Create AI agents with your voice — each one gets a custom, task-aware workspace designed for what it does.

Gumball is a mobile app where non-technical users describe what they need in plain language, and the app builds a custom workspace for it. Not a chatbot. Not a prompt box. A real, purpose-built UI that adapts to the task — housing trackers get price charts and signal cards, tweet composers get draft editors and tone pickers, research agents get source lists and metrics tables. The UI is the output.

An honest mirror for strength training — recruitment-weighted muscle heatmap, share-ready session cards, PR detection. No streaks, no gamified noise. → Details & TestFlight

Most lifting apps either copy guided-workout patterns (one set at a time, hidden inputs) or pile on streaks and badges to keep you swiping. HyperFit does neither: every set's weight and reps fields are visible at once, ghost-value placeholders show what you lifted last time, and the only "celebration" is a recruitment-weighted body silhouette filling in across front and back as you log. Anthracite base, strava-orange for active lifts, robinhood-green for session-volume gains.

A deep-work timer that doesn't lie about whether you focused. Five reaction-time taps before, five after — the chart tells the truth. No streaks. No celebration. Just the meter.

The average focus session in 2026 is 13 minutes 7 seconds — a three-year low. Most timers count minutes and reward presence. Brevn measures whether your brain was actually on. Each session is bookended by a thirty-second psychomotor-vigilance check (five reaction taps + a 1–5 alertness scale). The result screen shows the pre/post delta, a 0–100 quality score, and a one-line read in honest English: “Sharper after than before.” “Mixed signal — improved on one measure, not both.” “You finished, but you were sharper before. Worth noting why.” After three sessions, Trends surfaces personal patterns — sharpest hour, sweet-spot duration, whether breaks help you. Personal n=1 only. No stranger’s averages.