Full-Stack Product Engineer · Founder · ML & Scientific Software
Tymofii Hrynchuk
I build products end to end — architecture, backend, databases, UI, deployment, and the people using them on the other side. Six mobile apps, ten-plus web platforms, some machine learning, and a few hundred drones.
In short
I design, build, ship and grow products end to end — architecture, backend, databases, UI, deployment, and the users on the other side.
Self-taught full-stack product engineer and founder, five-plus years in. Six mobile apps and ten-plus web platforms launched independently; several have been live over two years with paying users.
Alongside the product work: a bachelor thesis training sixteen LSTM variants on a 3.5M-text FinBERT pipeline, and enough hardware to have assembled several hundred UAVs by hand.
- 6
- Mobile apps shippediOS, Android, one macOS
- 10+
- Web platformsSeveral live 2+ years
- 10,000+
- UsersAcross shipped products
- 16
- LSTM variants trainedControlled ablation study
Selected work
Shipped, not prototyped.
Every one of these went out to real users under my own name — including the billing, the store listings and the support.
Case — 2025 · iOS · Android · macOS
Flari
AI day-planning & screen-time app
My most technically complex app. An AI assistant turns a spoken day into a full structured plan, with real-time state synchronised across several interfaces and a stack of third-party integrations. The only product I have shipped with a native macOS build.
Won IdeasLab (UCU) — $25,000 in credits
- React Native
- SwiftUI
- RevenueCat
Bachelor thesis — NaUKMA
Instrumental Analysis and Forecasting of Financial Market Behaviour using NLP and LSTM
A controlled ablation study: four specialist LSTM models — return regression, directional classification, reversal detection, pump/dump detection — across four feature configurations, giving sixteen trained variants over four years of BTC/USDT data.
- 01Multi-source collection and feature engineering
- 023.5M+ deduplicated texts through a FinBERT sentiment pipeline
- 03Optuna hyperparameter optimisation
- 04Strict chronological evaluation, no look-ahead
- 05Next.js dashboard served by a Python REST API
- Pump/dump accuracy
- 69%82%
- Directional event F1
- 29%44%
- Python
- TensorFlow/Keras
- Optuna
- HuggingFace
- pandas
- scikit-learn
Hardware
Drones, printed frames, and a few hundred assemblies
- 2025
Lab418 — UAV Engineer & Technician
Helped design, assembled, tested and integrated several hundred UAV systems. Hands-on hardware build plus hardware/software systems integration.
- 2025
3D-Printed UAV — pre-diploma research
Designed, printed, assembled and flight-tested FPV drones with most frame parts printed, then compared flight performance, structural strength and noise across frame designs. Conclusion: printed frames are a viable, cheaper, customisable alternative to composites.
- 2026
EDTH Munich — 1st place
Built a drone friend-or-foe identification system from scratch in 48 hours, two-person team, at the European Defense Tech Hackathon.
Stack
Comfortable owning any part of it.
Currently getting familiar with Kubernetes, Helm, and distributed data / data-lake architecture.
Languages
- TypeScript
- JavaScript
- Python
- Swift
- SQL
- HTML/CSS
- Kotlin (basic)
Frontend & Mobile
- React
- Next.js
- React Native
- SwiftUI
- Three.js
- Tailwind CSS
Backend & Data
- Node.js
- REST APIs
- Prisma
- PostgreSQL
- SQLite
- Offline sync
- NextAuth
- Clerk
Machine Learning
- TensorFlow/Keras
- LSTM & time series
- Optuna
- HuggingFace
- pandas
- NumPy
- scikit-learn
Cloud & Infra
- AWS (GPU compute)
- Docker
- Vercel
- Cloudflare
- Git CI/CD
- RevenueCat
- App Store & Play
Hardware
- FPV/UAV design
- 3D printing
- AutoCAD
- Systems integration
- Ukrainian — Native
- English — C1
- French — Committed to learn
Awards & education
Competitive by habit.
BSc, Computer Science
National University of Kyiv-Mohyla Academy
2026 · Diploma 91/100
- Mar–May 2026
Winner — IdeasLab (UCU) accelerator
Selected and awarded $25,000 in cloud credits.
- 2026
1st place — EDTH, Munich
European Defense Tech Hackathon. Drone IFF system in 48 hours.
- 2024
Finalist — ReactorX accelerator
Top 20 of 400 applicants.
- 2017–22
National olympiad medallist
Physics, astronomy and mathematics — regional and town level, repeatedly.