Currently building Clavix · Based in the US
Building
software,
one thing
at a time.
Up at 4:30 most mornings. Right now I'm getting Clavix through App Store review. Before that was VarFoot, a World Cup predictor, and a health app. There's always something.
Sansar Karki
About me
I build iOS apps and web products, mostly by myself. The 4:30 AM thing is real. That's just when it's quiet and I can actually think.
I play soccer, so VarFoot started because the existing training apps were useless for high school players. Clavix takes up most of my time now. It filters financial news down to what actually moves the stocks you hold, and sends you a morning digest instead of a firehose.
I like working on things that have a real reason to exist. If I'm annoyed by something, I build a fix for it.
Stack
- SwiftUI & Swift
- Python
- Next.js & TypeScript
- Supabase & PostgreSQL
- Docker & Linux
What I'm building
-
01
Clavix
Filters financial news to what moves the stocks you hold, scores your portfolio's downside risk, and sends a morning digest. In TestFlight, App Store review submitted.
SwiftUI · FastAPI · Supabase · PostgreSQL · Docker
Clavix
Most financial news has nothing to do with what you actually own. Clavix filters it to what does. Connect your holdings and the app shows only the news that moves your stocks, scores your portfolio's downside risk, and sends a morning digest so you're not reading 40 headlines to find the two that matter.
Backend is FastAPI running on a self-hosted VPS. Database is Supabase/Postgres. The iOS app is SwiftUI with in-app purchases through StoreKit 2. The news scoring pipeline matches tickers, weights by sentiment, and ranks by position size.
In TestFlight now. App Store submission is in review.
-
02
VarFoot
Training app for high school soccer players who want to make varsity. You do an assessment, see where your numbers sit against varsity benchmarks, and get a week-by-week plan.
Next.js · TypeScript · Supabase · Gemini AI
VarFoot
I play soccer. The training advice for high school players online is mostly too vague to do anything with. VarFoot gives you an assessment, compares your numbers to varsity benchmarks, and builds a week-by-week plan from there.
It also handles food logging with macro tracking, shows your progress over time, and has an AI coach built on Gemini. Built as a PWA so it runs on any device without going through an app store.
-
03
WC2026 Predictor
Dixon-Coles bivariate Poisson stacked with XGBoost, built from public data. Beat bookmaker accuracy on 2018 and 2022 results. Live tool checks the model's odds against Kalshi and flags gaps.
Python · XGBoost · Flask · Dixon-Coles · Kalshi API
WC2026 Predictor
Dixon-Coles bivariate Poisson stacked with XGBoost, built entirely from free public data. On 2018 and 2022 World Cup results it beat bookmaker accuracy.
There's a live tool that checks the model's fair-value odds against Kalshi in real time and flags where the gaps are. A no-vig anchor from sharp markets vetoes signals that don't hold up. The model tracks its own accuracy and runs a Kelly-sized $1,000 mock portfolio. Paper money only.
Model output screenshots coming soon -
04
Concord
SwiftUI health app that reads from HealthKit and generates a report you can actually understand, with home-screen widgets. Report generation works. Still building the rest.
SwiftUI · Swift · HealthKit · WidgetKit
Concord
SwiftUI health app that reads from HealthKit and turns it into a report you can actually understand. The idea is to surface what your health data means without making you dig for it.
Report generation and home-screen widgets are working. The rest is still in progress, picking it back up once Clavix ships.
Screenshots coming soon
Say hi
Working on
something
interesting?
I'm not looking for work, but I'm happy to talk about what you're building. Drop me an email.