Experiments | Michal Kaszubski

Thesis bar chart of uncertainty-error AUROC, with error bars, for deep ensembles, MC dropout and test-time augmentation. ResNet-18 scores 0.64 to 0.69 and ResNet-50 0.60 to 0.65. The DenseNet-121 deep ensemble reaches only 0.52, just above the dashed chance line at 0.50.

MIQA prostate DWI

Research, 2025 to 2026

MSc research at UCL on automated detection of geometric distortion in prostate diffusion-weighted MRI, and on whether a model can tell when its own judgement is unreliable.

Role: MSc researcher, UCL (thesis author)
Built with: PyTorch, DenseNet-121, Deep ensembles, Selective prediction
Evidence: Research evidence
Result: Three-seed DenseNet-121 ensemble: mean AUROC 0.819, SD 0.049 across five patient-level folds, up to 0.17 AUROC over classical radiomic baselines. Retrospective triage simulation: 83.6% retained-case accuracy at 70% coverage (thesis, 2026).

Inspect the work

The AI Zuzi Socials Manager headline on a dark blue ground, with the line Create. Schedule. Automate. and a description of an all-in-one platform for creating, scheduling and distributing posts across channels.

An AI social-media management and content platform that I engineer for a client company through NeuralTake. Django, React and PostgreSQL, with seven third-party integrations, generation, payments and multi-factor sign-in. About 50 business users.

Role: Lead engineer through NeuralTake: back end, front end, integrations, security, database migration and deployment
Built with: Django, Django REST Framework, React, PostgreSQL
Evidence: Production deploymentUser adoption
Result: Used by about 50 business users, as stated in September 2026; a zero-data-loss live migration from SQLite to PostgreSQL; a security audit, encryption at rest, MFA and GDPR-aligned deletion before strangers were given accounts.

Inspect the work

NeuralKite

Products

A live acquisition-intelligence product for fragmented UK private markets. At its centre is an email discovery and verification engine that grades every address and keeps its reasons. Relevant to AI engineering, private-equity and finance roles.

Role: Product engineer, end to end: verification engine, data layer, scrapers, Claude agents, application and sending
Built with: Next.js App Router, TypeScript, Anthropic SDK, PostgreSQL
Evidence: Production deployment
Result: Deployed and publicly accessible at neuralkite.com, verified 15 September 2026. Its verification engine grades every address into a risk tier with a written rationale. Customer and usage figures are not published.

Inspect the work

ChefBot

Products, 2024 to present

A cooking assistant that turns whatever is in the fridge into a meal. It began as a GPT (25,000+ conversations, rated 4.7 from 100+ ratings) and continues as a ChatGPT plugin with 18 skills.

Role: Designed the retrieval, prompt logic and multi-step tool flows; orchestration on GPT, not model training
Built with: ChatGPT plugin (skills), Prompt logic, Multi-step tool flows, Image input
Evidence: User adoptionProduction deployment
Result: 25,000+ conversations and a 4.7 rating from 100+ ratings as a GPT (store listing, screenshot September 2026; conversations, not people); continues as a plugin since October 2026.

Inspect the work

NeuralTake

Products, 2024 to present

The AI consultancy I founded in London in 2024. It automates manual workflows for professional services firms: a two-week audit, then fixed-fee builds that live in the client’s own repository. I sell, scope, build and support the work.

Role: Founder and AI engineer; discovery, scoping, architecture, delivery, onboarding, support and roadmap
Built with: Python, Django, TypeScript, Next.js
Evidence: Commercial validationProduction deployment
Result: Signed B2B contracts delivered end to end, among them AI Zuzi, a client platform now in production. Clients and contract values are not published.

Inspect the work

RSNA knee MRI challenge

Research, 2026

An entry to the 2026 RSNA Knee Abnormality Detection competition on Kaggle, now closed: radiology reports taught an EfficientNetV2 and DINOv2 ensemble to read knee MRI from the images alone. It reached 0.941 on the public leaderboard by itself.

Role: Competitor, solo: DICOM data engineering, weak-supervision pipeline, model design and training, submission engineering
Built with: PyTorch, EfficientNetV2, DINOv2, timm
Evidence: Public benchmark
Result: 1,849th of 6,425 teams (private 0.942, public 0.958) from a blend built on the strongest public notebook. My own ensemble reached 0.941 on the public leaderboard alone, 1 September 2026. No medal.

Inspect the work

Cash Nova

Experiments, 2026

A fictional producer whose tracks I generated programmatically through the Eleven Music API. The pipeline covered music, voices, sound matching and sequencing, with six albums and 52 tracks on ElevenMusic in 2026.

Role: Built and ran the automated generation pipeline
Built with: Eleven Music API, ElevenLabs voices
Evidence: Creative experimentAudience reach
Result: 52 tracks generated through the Eleven Music API and released as six albums.

Inspect the work

Two vertical video thumbnails side by side: an animated celery plant with a face standing in a bowl of water, captioned in English on the left and in Polish on the right.

CleverFacts and Ogarnik

Experiments, 2026

An automated pipeline that turns one short-form educational concept into an English and a Polish video through generation and dubbing APIs, measured with the full YouTube Analytics history of both channels: 148 videos, 73 matched pairs.

Role: Built and ran the automated production, dubbing and analytics pipeline for both channels
Built with: Google Veo (Vertex AI), ElevenLabs API, YouTube Analytics API v2, YouTube Data API v3
Evidence: Audience reachCreative experiment
Result: 727,455 views across 148 short-form videos in English and Polish (YouTube Analytics, 11 March to 19 September 2026). Views, not unique viewers. Across 73 matched pairs, English and Polish results correlated only loosely (rank correlation 0.46).

Inspect the work