École Polytechnique Fédérale de Lausanne
4 days ago
Postdoc in Machine-Verified LLM Inference, Formal Verification, and Computer Systems at EPFL École Polytechnique Fédérale de Lausanne in Switzerland
Degree Level
Postdoc
Field of study
Computer Science
Funding
Postdoctoral position funded by an ARIA-funded collaboration between EPFL and Imperial College London. Full-time 100% contract, 1 year renewable depending on project duration. Travel funding is provided for collaboration between Lausanne and London and for conferences; access to frontier AI models and high-performance compute is also offered.
Country
Switzerland
University
École Polytechnique Fédérale de Lausanne

How do I apply for this?
Sign in for free to reveal details, requirements, and source links.
Apply for this position
Keywords
Suggested positions
About this position
EPFL (École Polytechnique Fédérale de Lausanne) is advertising a Postdoc: Keystone Project (Machine-Verified LLM Inference) in Switzerland. The project sits at the intersection of formal verification, interactive theorem proving, computer systems, ML infrastructure, and LLM inference.
The Keystone project aims to advance machine-checked verification for systems software and AI-assisted proof engineering, with possible research directions including GPU kernel semantics and verification, verified inference coordination layers (batching, KV-cache management, scheduling), and agentic AI workflows for specification autoformalization, proof generation, and proof repair. The role is part of an ARIA-funded collaboration between EPFL and Imperial College London, with regular visits to London expected.
Applicants should have a PhD in computer science or a closely related field, or be nearing completion. Relevant experience includes Rocq/Lean/HOL/Isabelle, GPU programming or semantics, compilers, concurrency, ML systems, and strong software engineering skills in Python, C++, OCaml, or functional languages. A publication record, strong English communication, independence, creativity, and motivation for research at the AI/formal methods interface are emphasized.
This is a full-time postdoctoral contract (100%), starting on 01.12.2026 or by agreement, for 1 year renewable depending on project duration. Funding includes travel support for Lausanne–London collaboration and conferences, plus access to frontier AI models and high-performance compute for proof-assistant workloads.
Apply only via the EPFL online platform. Submit a brief cover letter, a single PDF containing your CV with publication list and research statement, and contact details for 3 referees. For further information, contact Nate Foster at [email protected].
Funding details
Postdoctoral position funded by an ARIA-funded collaboration between EPFL and Imperial College London. Full-time 100% contract, 1 year renewable depending on project duration. Travel funding is provided for collaboration between Lausanne and London and for conferences; access to frontier AI models and high-performance compute is also offered.
What's required
PhD in computer science or a closely related field, or nearing completion. Background in formal verification, programming languages, systems, or ML. Research experience in interactive theorem proving (Rocq, Lean, HOL, Isabelle, or similar), GPU programming or semantics, compilers, concurrency, or ML systems is preferred. Strong software engineering skills across multiple languages such as Python, C++, OCaml, or functional languages. Publication record relative to career stage, excellent English communication skills, independence, creativity, and strong motivation for AI and formal methods. Experience with LLM-based tools for code or proof generation is a plus.
How to apply
Apply through the EPFL online platform only. Prepare a brief cover letter (up to 2 pages), one PDF containing a CV with publication list and a research statement (up to 3 pages), and contact details for 3 referees. For questions, contact Nate Foster by email.
More information can be found here
Ask ApplyKite AI

How do I apply for this?
Sign in for free to reveal details, requirements, and source links.