AI Engineer (RL Environments)

Hashlist; Helsinki; On-site; EUR 80,000 - 120,000; Full time

Benefits offered by this job

Competitive compensation

Meaningful equity

Central office in Helsinki

Work with model providers

Job summary

Hashlist expands its AI research team in Helsinki to develop domain-specific RL environments for constrained embedded programming and enterprise workflows. You will build and automate platforms for RL environments and simulate worlds to reveal model weaknesses.

You will translate training objectives into concrete data and evaluation specs, construct the reward layer, and run large-scale sandbox rollouts while collaborating with OEMs and AI Labs on fine-tuning challenges.

Qualifications

• Production ML depth with Python and PyTorch for RL training.
• End-to-end RL environment built and model trained.
• Experience with containerization tools like Docker and scaling systems.
• Strong communication and motivation.

Responsibilities

• Build and automate our platform for creating RL environments.
• Construct simulated worlds and expose model failure modes.
• Turn AI objectives into concrete data and evaluation specs.
• Build the reward layer and mitigation for reward hacking.
• Run rollouts at scale with hundreds of sandboxed attempts.
• Fine-tune open-source models and export datasets.
• Communicate with OEMs and AI Labs on research questions.

Skills

Python

PyTorch

Reinforcement learning

TRL

OpenRLHF

LoRA fine-tuning

vLLM

SGLang

PPO

Docker

Tools

Docker

Job description

Would you like to operate at the frontier of AI evaluation, post-training, and model improvements?

We are now expanding our core Hashlist AI research team to support the creation of domain-specific RL environments for constraint-based embedded programming & complex enterprise engineering workflows.

What you will do:

• Build and automate our platform for creating RL environments
• Construct simulated worlds and explore data shapes that expose meaningful model failure modes across the embedded coding domain & related enterprise workflows
• Turn AI training objectives into concrete data and evaluation specifications
• Build the reward layer + reward-hacking mitigation
• Run rollouts at scale. Hundreds of sandboxed attempts per task in parallel
• Fine-tune open-source models: Before-and-after fine-tunes, failure reports, and dataset exports
• Communication with our clients (OEMs & AI Labs) on specific research or fine-tuning questions they have

Skills needed:

• Production machine learning depth. Python and PyTorch, reinforcement learning training with TRL, verl, SkyRL or OpenRLHF, PPO, GRPO or DPO, LoRA fine-tuning, and inference with vLLM or SGLang.
• Training or evaluation systems, including at least one RL environment you built end-to-end and trained a model against. Comfortable with the infrastructure around it, e.g Docker, or similar containerization tools, to design and monitor systems at scale
• Experience with harness & agentic optimisation for evals
• Strong familiarity with common reinforcement learning algorithms and methods, especially with respect to post-training LLMs
• High level of personal drive, motivation, and good communication skills.

Bonus:

• You have published an environment, benchmark or evaluation harness we can look at.
• You have worked with embedded, safety-critical or other physical engineering software.

Company benefits

• Competitive compensation + meaningful equity
• Central office in Helsinki
• Be a part of a quickly scaling tech company working directly with model providers