Senior ML Research Engineer

Agent Learning and Verifiable Reasoning

AI4I · PRAXISTurin, ItalyFull time · Two years

Call upcoming

Develop agents that learn to reason about contracts and produce the evidence needed to act under them.

Your role

You will advance the capabilities of agents that learn from reasoning, verification and interaction. The scientific objective is to develop models that become better at capturing requirements, constructing evidence and revising their decisions, while understanding whether these improvements transfer to unfamiliar contracts, tools and tasks. The role offers the opportunity to shape a model development agenda at the intersection of agent learning and formal reasoning.

You will take technical and scientific ownership of model improvement, combining practical training with the freedom to propose methods and lead experimental studies. Working alongside postdoctoral researchers and specialists in formal methods, you will lead the investigation of learning processes and turn new methods into reproducible experiments. You will lead research contributions and reusable releases alongside sustained ownership of the learning infrastructure.

Within PC2, directions include fine tuning, reinforcement learning with verifiable rewards, and learning from iterative or recursive processes for solving problems. These processes include decomposing obligations, repairing failed proofs, refining contracts after counterexamples and changing plans after rejected actions. You will investigate how these processes generate training signals and datasets, and which approaches improve the capabilities that the programme needs.

You will own the training and evaluation infrastructure that makes these investigations repeatable, adapting methods and diagnosing failures in models and feedback mechanisms. A central question is how improvements on a checking signal relate to useful task completion and faithful intent capture: these outcomes should not be conflated. You will work with the relevant researchers to establish informative comparisons and held out tests; prior expertise in formal methods is not required.

Your contribution to the field

Your contribution will advance learning methods that turn verification and iterative reasoning into stronger agent capabilities. You will develop and evaluate approaches that use feedback from proof repair, contract refinement and planning, establishing how their benefits transfer to unfamiliar tasks. Research methods, model releases and reproducible training systems will give your work an identifiable place at the intersection of post training, agent learning and formal reasoning.

An initial research or development direction

An example starting point is to study how training on one recurring reasoning process, such as proof repair or contract refinement, changes performance on new tasks. Strong unadapted baselines and controlled comparisons will establish where the learning method adds value and guide further model development.

Essential background

Substantial practical experience training or adapting machine learning models, including fine tuning or closely related post training work.
Strong implementation and experimental skills, with the ability to diagnose training behaviour, evaluate alternatives and make reproducible improvements.
Ability to lead a model development project from task and data design through training, evaluation and analysis.
Relevant technical training or equivalent professional experience. A PhD is not a prerequisite for this engineering profile; formal appointment eligibility will be specified in the final call.

Additional relevant experience

Reinforcement learning with verifiable rewards, agent training, proof or program synthesis, or learning from search and repeated feedback.
Distributed training, efficient fine tuning, experiment management and data pipelines.
Formal methods, dependent type theory or interaction with proof assistants. These are useful additions, not entry requirements.

Candidates with depth in a relevant area and an interest in connecting it to others are encouraged to apply. Evidence of your contribution can include research papers, formal developments, substantial software or a clear account of technical work that cannot be shared publicly.

Research environment and development

Your primary supervisor will be Carlo Ciliberto. PC2 is a joint effort involving PRAXIS at AI4I, the REAL team at Bocconi and STAMP at Inria. The role requires close collaboration with the other participating groups, including opportunities for extended visiting periods at their laboratories, bringing your work into close contact with complementary expertise in machine learning, formal methods and systems.

The position is based in Turin, Italy. Remote working arrangements can be agreed when working from another laboratory in the PC2 collaboration. The timing and practical arrangements for research visits will be discussed with you and the participating hosts.

Computational resources. AI4I's HPC infrastructure provides approximately 150 NVIDIA GPUs (80 Blackwell B200 and 68 Hopper H200) and 5 PB of storage, supporting this project in agentic AI training, testing and validation.

Application materials

Please prepare:

Your CV, including relevant research or technical experience.
A short statement of fit and interests, explaining what attracts you to the role, the experience you would bring and directions you would be interested in exploring. A full research proposal is not required.
One or two representative contributions, such as papers, formal developments or software. Briefly explain your own contribution. An account of work that cannot be shared publicly is welcome; a public repository is not required.

Submission instructions and the application deadline will be published here when the call is finalised.