Sebastian BoehlerResearch engineer & founder

Tübingen, Germany / Research & practice

An idea is
a starting point.

I build AI systems and the environments they learn in. Working across simulation, autonomous systems, markets, and learning.

01 / The shape of learningHeight = loss

Gradient descent

Small steps. A different perspective.Illustrative objective · θ₁, θ₂

Selected work

01 — 04

01

Autonomous systems

Sunderlabs / FlightRL

Training compact policies before they reach real hardware.

High-throughput native environments, privileged teachers, edge-shaped policies, telemetry, and staged gates.

02

Markets & infrastructure

HB Capital

Keeping market claims attached to their evidence.

Market structure, positioning, macro context, native research tooling, and read-only account context.

03

Language models

Dialogue research

Testing whether smaller language models can predict what users say next.

QLoRA next-turn prediction with multi-step dialogue rollout evaluation.

04

Learning systems

LecturePilot

Turning private course material into controlled learning workspaces.

Professor-owned sources, enforced unlocks, typed tutor capabilities, and learner-owned work.

Notes & observations

All writing ↗

Why Brains and LLMs Both Become Modular

XGBoost and the Map of Machine Learning Approaches

Parameter Golf and Distributed Autoresearch

Contribution arc

A quiet record of sustained public building from 2017 onward.

2026
7,333 contributions
2025
1,798 contributions
2024
1,145 contributions
2023
256 contributions
2022
524 contributions
2021
554 contributions
2020
574 contributions
2019
313 contributions
2018
1 contribution
2017
2 contributions

A little context.

I moved from self-taught builder and backend engineer to research engineer and founder. Today I work through Sunderlabs and HB Capital while studying computer science at Tübingen.