fafula.com

Aleksander Fafuła, PhD

Working in AI since 2007.

scroll

01Profile

I build AI and data systems, from infrastructure to the decisions they support. My work spans financial markets, banking and industry. I hold a PhD in quantitative behavioural finance and study how AI models behave when making decisions under uncertainty. I also founded Indexis, an independent, public-interest analytics boutique. FreeBSD was my main OS as a kid; the terminal has been home ever since. I also spent a few years as a touring rock drummer.

02My initiatives

FunduszAI

funduszai.pl

A public experiment in AI decision-making on the Warsaw Stock Exchange. AI agents analyse market data, debate each stock and record a position with a confidence level. Their calls are tracked over time, making the system a test of how AI behaves under real market conditions. Not investment advice.

Indexis

indexis.pl

An independent, public-interest analytics boutique I founded to turn data into clear indicators and analysis for public debate. Built on proprietary data infrastructure and transparent, repeatable methods.

Kierunek Innowacje (Direction: Innovation)

kierunekinnowacje.pl

My Polish-language podcast on technology and business, available on Spotify, YouTube and Apple Podcasts. Conversations about how people put AI and new technology to use in their organisations.

AI Drummer

aidrummer.io

An experimental interface for controlling real drum hardware through conversation. It brings together my background in music and AI: describing a sound or a setup instead of navigating menus.

03Research & open source

Abliteration Is Not a Scalpel

Read the preprint on arXiv

My study of how removing refusal behaviour from language models changes their decisions on tasks where they would not refuse in the first place. It examines shifts in optimism, confidence and reasoning across model families. Preprint, 2026.

OneSpark

Code and validation on GitHub

An open-source adaptation for running an abliterated Qwen3.8-Flash-Next model on a single NVIDIA DGX Spark. Built on community work, with a reproducible setup and published validation logs. A practical experiment in running large models locally.

04Speaking

DevAI by Data Science Summit

19 October 2026 · Online · Talk in Polish

A talk on what goes wrong in production AI agents: decision traps, plausible explanations and silent changes in model behaviour. Drawing on FunduszAI and my research on how weight modifications affect model decisions.

05Background

PhD in quantitative behavioural finance, awarded with distinction.

Doctoral dissertation recognised in the national NTIE competition in economic informatics, 2014.

Two-time winner of Digital Ars, Poland's first AI-art competition (DigitalPoland) — the open-category piece, "AI painting with the drums", a GAN that paints in real time to live drumming.

06Experience

  • Principal AI & Data Architect (2025–present). Current engagements are under NDA.
  • Chief Data Officer — Chabre, data & AI consulting (2023–2025). Client projects in the Gulf.
  • Director, Data Architecture & Analytics — Diversey, a global US/Dutch chemical and manufacturing company (2023). Predictive maintenance for a 1,200-robot fleet.
  • Global Enterprise Solutions Architect — British American Tobacco, London (2020–2023). Data-science architecture for revenue growth and marketing, rolled out across international markets.
  • Head of AI & Data Science Infrastructure — Publicis Groupe, advertising & media (2018–2020). Built the group's geo-analytics for campaign targeting.
  • Data Analytics & Big Data, Manager — PwC Poland, consulting (2015–2018). Built a big-data platform for a global food retailer that lifted average basket size up to 23% — a Cloudera customer story. Also built M.A.I.A., PwC's AI recruitment assistant, and big-data systems in telecom and banking.
  • Chief Data Scientist — MarketPsych, US (2011–2015). The behavioural-finance firm behind the Thomson Reuters MarketPsych Indices — a market-sentiment data feed used on Wall Street by hedge funds and trading desks. Research I produced there is cited in Richard Peterson's Trading on Sentiment (Wiley).
  • Assistant Professor, Lecturer & Researcher — Wrocław University of Economics, Kozminski University (2007–2022). Teaching AI and data science.

07Contact

Privacy & analytics

This site uses Cloudflare Web Analytics for page views, referral sources and page performance. The service does not use analytics cookies, local storage or persistent visitor identifiers. Cloudflare states that incoming IP addresses are discarded at its nearest data centre. See Cloudflare's data collection details. This site's analytics respects Do Not Track and Global Privacy Control.

Fonts are served with the site. Videos and social profiles are linked rather than embedded. For questions about privacy, email aleksander@fafula.com.