AI Researcher - Post-Training (f/m/d)

Sonar Source
Germany
Workplace: OnsiteFull timeEUR 90,400 - 166,800 annuallyFunction: Research & Scientific (R&D)Experience: 4+ yearsSkills: ["Communication","Explaining complex topics","Research delivery","Prototype-to-product mindset","Collaboration"]

Develop enterprise-grade coding agents and models by combining Sonar’s static analysis expertise with leading large language model post-training techniques. Drive research from hypothesis and experiments through rapid prototypes and into productized outcomes. Contribute as a subject matter expert on policy optimization, verifier-based RL frameworks, supervised fine-tuning, PEFT, and preference/safety alignment, collaborating across researchers, research engineers, MLOps, and engineers. Explain complex topics clearly to technical and non-technical audiences.

Loading

Loading job details...

Preparing the role view and application actions.

FursaFursa
Sonar Source
Sonar Source
1 week ago

AI Researcher - Post-Training (f/m/d)

✓ Verified Job

Canonical indexed version, validated from employer's careers page.

Source: Company careers pageValidated by: Fursa AI
Last checked: 23 hours agoStatus: Live

Job Summary

Develop enterprise-grade coding agents and models by combining Sonar’s static analysis expertise with leading large language model post-training techniques. Drive research from hypothesis and experiments through rapid prototypes and into productized outcomes. Contribute as a subject matter expert on policy optimization, verifier-based RL frameworks, supervised fine-tuning, PEFT, and preference/safety alignment, collaborating across researchers, research engineers, MLOps, and engineers. Explain complex topics clearly to technical and non-technical audiences.
Location: Germany
Workplace: Onsite
Employment Type: Full time
Job Function: Research & Scientific (R&D)
Seniority: Mid level

Key Responsibilities

  • •Innovate and develop next-generation solutions for enterprise-grade coding agents and models.
  • •Harness Sonar’s static analysis experience and apply it with leading LLM post-training techniques.
  • •Work on policy optimization, verifier-based RL, supervised fine-tuning, synthetic data/curation, PEFT, and preference/safety alignment.
  • •Translate research prototypes into cutting-edge products by collaborating with researchers, research engineers, MLOps, and engineers.
  • •Stay up to date with LLM and agentic developments and explain complex technical details to technical and non-technical audiences.

Pay and Benefits

Salary: EUR 90,400 - 166,800 annually

Key Requirements

  • •Advanced academic background (Master’s or PhD) in Computer Science, Machine Learning, or a related quantitative field.
  • •4+ years industry experience in machine learning with a solid understanding of modern software engineering practices and tools.
  • •Fluency with Python and core ML frameworks; experience with Rust or SonarQube languages (C#, C++, JS/TS, Java) is a plus.
  • •Expertise in LLM post-training, including policy optimization (e.g., GRPO, PPO), verifier-based RL frameworks (RLVR), supervised fine-tuning (SFT), synthetic data/curation, and parameter-efficient fine-tuning (PEFT) with preference and safety alignment.
  • •Experience driving research projects from valuable findings and prototypes through product conversion, with excellent communication skills in English.
Experience:4+ yearsMachine learningAIAgentic software developmentLLMsSoftware engineering
Education:
Skills:CommunicationExplaining complex topicsResearch deliveryPrototype-to-product mindsetCollaboration
Languages:English
Tech Stack:PythonRustC#C++JS/TSJavaGRPOPPORLVRSFTPEFT

Company Brief

Sonar Source
Builds static code analysis and continuous inspection tools (SonarQube, SonarCloud, SonarLint) that identify bugs, vulnerabilities, and code smells across multiple languages to help teams improve code quality and maintainability.
Industry: Developer Tools
Company Size: Large (251 to 1,000 employees)
Growth: Established Company
Headquarters: Geneva, Switzerland
Founded: 2008
WebsiteLinkedIn