What You Will Bring
Deep Post-Training Expertise: Hands-on experience with post-training models for specific applications (SFT, RLHF, RLAIF, Reward Modeling, Knowledge Distillation, etc)
Strong Architectural Foundations: Deep understanding of Transformer architectures (attention mechanisms, positional encodings) and ML systems. And knowing when to use which model (sometimes it's simpler models that win!)
Large-Scale Training: Experience with distributed training frameworks and optimizing training jobs on GPU clusters.
High Velocity & Ownership: You thrive in ambiguous environments, learn quickly, and have a bias toward action. You are comfortable shipping in rapid cycles typical of early-stage startups.
Technical Stack: Proficiency in Python, PyTorch. Familiarity with the modern open-source LLM stacks (HuggingFace, Vertex, Vercel, etc.).
(Healthcare experience is preferred but not strictly required if you have exceptional ML fundamentals.)
What You’ll Work On
Architect the Post-Training Stack: Lead the design and execution of alignment pipelines (SFT, RLHF, RLAIF) that bridge the gap between "exam-passing" models and "clinically useful" systems.
Leverage Proprietary Data: Utilize our proprietary and open source medical datasets to fine-tune models on edge cases that generic models miss.
Novel Technique Experimentation: Research and implement cutting-edge post-training methods to optimize model performance, aiming for improvements in calibration and reliability critical for healthcare.
Safety & Evaluation: Build rigorous evaluation frameworks (LLM-as-a-judge, benchmarks) to detect hallucinations, ensure clinical correctness, and guarantee safety before deployment.
Strategic Collaboration: Work directly with the co-founders to define the research roadmap and platform strategy.
Bonus Points
Research Track Record: Published research in high-impact journals or top-tier ML/AI conferences (NeurIPS, ICML, ICLR, CVPR, ACL).
Top Lab Experience: Background working or interning at top research labs (e.g., FAIR, DeepMind, OpenAI, Google DM, MSR, Stanford/CMU/MIT labs).
Domain Expertise: Experience dealing with multimodal health data, clinical reasoning, or safety-critical ML systems.
Entrepreneurial Spirit: You have founded a company, built early-stage products, or enjoy the "zero-to-one" phase of building.
Skills Required
- Hands-on experience with post-training models (SFT, RLHF, RLAIF, Reward Modeling, Knowledge Distillation)
- Deep understanding of Transformer architectures, attention mechanisms, and positional encodings
- Experience with distributed training frameworks and optimizing training jobs on GPU clusters
- Proficiency in Python
- Proficiency in PyTorch
- Familiarity with modern open-source LLM stacks (HuggingFace, Vertex, Vercel, etc.)
- Ability to operate in high-velocity, ambiguous early-stage startup environments with strong ownership
- Healthcare or clinical ML experience
- Published research in top-tier ML/AI conferences or journals
- Experience at top research labs (FAIR, DeepMind, OpenAI, Google DM, MSR, top universities)
- Domain expertise with multimodal health data, clinical reasoning, or safety-critical ML systems
- Entrepreneurial experience (founded company or built early-stage products)
What We Do
TriFetch is a San Francisco-based healthcare AI startup that provides an end-to-end automation platform for independent specialty clinics and multi-specialty groups. The company automates critical administrative workflows, including patient calls, scheduling, referral processing, and prior authorizations. By reducing the administrative burden on staff, TriFetch enables healthcare providers to focus more on patient care while integrating seamlessly with existing EMR systems.
