Research, Mid Training

Thinking Machines Lab
San Francisco
Workplace: HybridFull timeUSD 350,000 - 475,000 annuallyFunction: Education & TrainingEducation: bachelorsSkills: ["Communication","Debugging","Problem-solving"]

Own late-stage mid-training for foundation models, shaping core reasoning capabilities through data mix design, synthetic data strategies, and quality uplift. Build and calibrate data pipelines for filtering, deduplication, verification, and rewriting, and develop quality classifiers and LLM judges to raise standards at scale. Iterate on mid-training recipes, evaluations, and training stability, while debugging distributed training and scaling methods to explore new approaches.

Loading

Loading job details...

Preparing the role view and application actions.

FursaFursa
Thinking Machines Lab
Thinking Machines Lab
14 hours ago

Research, Mid Training

✓ Verified Job

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

Source: Company careers pageValidated by: Fursa AI
Last checked: 4 hours agoStatus: Live

Job Summary

Own late-stage mid-training for foundation models, shaping core reasoning capabilities through data mix design, synthetic data strategies, and quality uplift. Build and calibrate data pipelines for filtering, deduplication, verification, and rewriting, and develop quality classifiers and LLM judges to raise standards at scale. Iterate on mid-training recipes, evaluations, and training stability, while debugging distributed training and scaling methods to explore new approaches.
Location: San Francisco
Workplace: Hybrid
Employment Type: Full time
Job Function: Education & Training
Seniority: Mid level

Key Responsibilities

  • •Own the data for each capability and knowledge area by sourcing, curating, and synthesizing training-grade datasets.
  • •Improve what the model knows via interventions such as targeted corpora, synthetic rephrasings, and knowledge-dense mixtures, and measure retention behavior.
  • •Build and operate an automatic quality pipeline by training/calibrating quality classifiers and LLM-based judges, plus verifiers for synthetic data.
  • •Develop and tune mid-training recipes, including dataset collections, training stages, annealing schedules, and hyperparameters, and assess downstream impact.
  • •Run an iterative loop of evaluations and debugging to make both metrics improve and ensure evals remain meaningful.

Pay and Benefits

Salary: USD 350,000 - 475,000 annually
Perks:Health InsuranceDentalVisionPaid LeaveParental LeaveRelocation

Key Requirements

  • •Proficiency in Python and familiarity with deep learning frameworks (PyTorch, TensorFlow, or JAX), with comfort debugging distributed training and scalable code.
  • •Bachelor’s degree or equivalent experience in Computer Science, Machine Learning, Physics, Mathematics, or a related field with strong theoretical and empirical grounding.
  • •Clarity in communication, with ability to explain complex technical concepts in writing.
  • •Strong grasp of probability, statistics, and ML fundamentals, including distinguishing real effects from noise and bugs.
  • •Experience owning training datasets for large models, including synthetic data pipelines and real user data (collection, curation, filtering, or mixture design).
Experience:Deep learningFoundation modelsData-centric AIHuman-AI collaborationDistributed training
Education:Bachelor's
Skills:CommunicationDebuggingProblem-solving
Tech Stack:PythonPyTorchTensorFlowJAX

Eligibility

Work Authorization:Sponsorship available.

Company Brief

Thinking Machines Lab
Develops enterprise AI solutions, custom large language models, and ML platforms to help organizations deploy intelligent applications. Services include data engineering, model development, and AI consulting for scale and production readiness.
Industry: AI & Machine Learning
Company Size: Medium (51 to 250 employees)
Growth: Growth Stage Startup
Headquarters: Mumbai, India
Website