A little about me
I’m Jesús Pérez Bazarot, 21, born and raised near Seville and now in my third year of Computer Engineering at the Universidad de Sevilla — with a year at the Beijing Institute of Technology in between. I work on the parts of AI that break: models that forget what they learned, models too large for the hardware they need to run on, benchmarks that quietly measure the wrong thing. Most of what I know I learned by building something, watching it fail in an interesting way, and rebuilding it. Along the way I've spoken at Telefónica, competed in NASA SpaceApps, been CTO of a small startup, and co-founded another. Martial arts, three or four times a week, taught me the part that no project teaches: how to keep showing up after losing.
What I do
I build benchmarks and the systems they measure. If a result can’t be reproduced bit for bit, I don’t trust it yet — including my own.
My current research asks a narrow question: when a world model learns a second task, what exactly does its transition model forget, and how do we measure that honestly? Building it taught me more about experimental hygiene than about world models. Seven debugging sessions turned up twenty-one bugs, one of which — a collapsed VAE posterior — invalidated an entire round of results and every conclusion I had drawn from them. Rewriting the paper from scratch was less painful than publishing something I knew was wrong.
Outside research, I write Rust and Python, deploy models on hardware that has no business running them, and keep a small set of tools I use daily. I’m aiming for a master’s in AI at MIT, and long term for work where machine learning is pointed at something worth conserving — marine ecology, for a start.
Off the screen
Experience
-
Independent AI/ML Research World models, continual learning, LLM benchmarking2025 — Present
-
AI/ML Developer OrgaAI — conversational AI, audio & transcription(under NDA)
-
Co-founder & Co-CTO ByTheWay Carpool — closed2024 — 2025
-
Chief Technology Officer NetKey — NFC networking hardware, closed2023 — 2024
-
Technology Speaker — innovaTE Telefónica — NB-IoT & sustainability track2023
Education
-
BSc Computer Engineering Universidad de Sevilla — Computer Engineering specialisation2023 — 2027 (expected)
-
Erasmus exchange Beijing Institute of Technology — CS & Big Data AnalysisSep 2025 — May 2026
Selected work
-
WMF Benchmark Catastrophic forgetting in world models — paper in preparation2026
-
HybridMamba-11 OpenAI Parameter Golf — submitted April 20262026
-
Perseo Personal AI assistant — Tauri 2, Rust, Gemini Live2023 — Present
-
Agentic Reasoning Traces 15 annotated traces, 142 steps, 9-type error taxonomy2026
-
ClassTranscriber 7-stage lecture pipeline — diarisation, Whisper, local cleanup2025
Achievements
-
Wanted Entrepreneurship Competition Winning team — venture and innovation execution2023
-
Debate Competition — finalist Provincial level2023
Toolkit
-
Languages Python, TypeScript, Java, C, C++, Rust, SQL, Assembly
-
Deep learning PyTorch, CUDA, Triton, torch.compile, BF16, quantisation, DDP
-
AI/ML Transformers, state space models, RAG, speech pipelines, diarisation, benchmarking
-
Backend & data FastAPI, React, Supabase, PostgreSQL, pgvector, Docker, Neo4j, WebSockets
-
Spoken Spanish (native), English (C1, Trinity 2020), Chinese (conversational)
