I post-train task-specific & agentic models.
Hi! I'm David Podolskyi βοΈ Software & ML Engineer @ Apple, where I post-train task-specialist models that replace frontier-model calls: soft-verified SFT, on-policy RL, reward design, evals. Previously: ML research & thesis @ Harvard Medical, President of TUM.ai, and five years of ML & data engineering across Infineon, BMW, and startups.
Outside of training runs: biology, physics, and maths, designing products people actually want to use, and an unreasonable amount of pickleball.
Explore more here
Selected work
Four things that show how I build.
Post-Trained Task Specialists @ Apple
Soft-verified-generation SFT per SERA and on-policy RL with SDPO, training small specialists that replace frontier-model calls at sub-dollar-per-fix cost. The task they learn: resolving internal issues into PR-ready code diffs, with a ModernBERT classifier fine-tuned on ~10K tickets (85% top-3 accuracy) routing the work.
RL from Scratch: PokΓ©mon TCG (Kaggle Silver)
Top 3% of 6,892 teams (Silver). One of the biggest Kaggle challenges ever. Agent: a policy/value transformer trained by behaviour cloning, then PPO against a self-play league, then per-deck LoRA specialists fused back into the base. At match time it plans with determinized tree search over a belief model, on 2 CPUs with no network.
Neural Rendering for Neurosurgery @ Harvard Medical School
My Bachelor's thesis: aligning what a surgeon's camera sees with preoperative brain scans, using NeRFs as differentiable scene representations for intraoperative registration.
TUM.ai & Community
Ex-President of TUM.ai for over two years, joining early and helping build it from a handful of students into one of Europe's best-known student AI initiatives: 300+ members, summits, hackathons, startup coaching, and industry projects. Also co-organized Europe's largest defense hackathon (400+ participants).
Work
- Company
- Apple
- Date
- Role
- Software & Machine Learning Engineer
- Company
- Apple
- Date
- Role
- Software & ML Engineer Intern
- Company
- UTUM, Venture Labs Software/AI
- Date
- Role
- Project Manager
- Company
- Harvard Medical School
- Date
- Role
- Machine Learning Researcher - Brain Surgery Imaging
- Company
- AISBACH
- Date
- Role
- Tech Consultant - ML & Software Engineering
- Company
- Infineon Technologies
- Date
- Role
- AI Solutions Engineer
- Company
- Expleo Group
- Date
- Role
- Data Engineer for IoT Systems
- Company
- TURTLE GmbH
- Date
- Role
- Data Engineer & Science Consultant
- Company
- AI Driller
- Date
- Role
- Junior Data Scientist & ML Engineer
Volunteering & Social Engagement
- Company
- TUM.ai
- Date
- Role
- President
- Company
- TUM.ai
- Date
- Role
- Industry Department - Team Lead & Mentor
- Company
- UnternehmerTUM & AppliedAI
- Date
- Role
- Decentralized Help Hotline for Ukrainian refugees
- Company
- TUM.ai
- Date
- Role
- Venture Department - Team Lead
- Company
- TUM.ai
- Date
- Role
- Events Organiser & ML Consultant













