AI/ML Scientist / Engineer — LLM Post-Training

eGov Foundation; Bengaluru; On-site; INR 2,500,000 - 4,500,000; Full time

Job summary

Sahai by eGovernments Foundation seeks an AI/ML Scientist to develop and validate a proof-of-concept model for clinical interviewing and diagnostic decision support. The role focuses on model research and development, not production deployment.

You will collaborate with clinicians, data specialists, and engineers to design experiments, post-train candidate models, and establish what works in Indian primary care.

Qualifications

• Strong ML research background with hands-on model development.
• Experience post-training open-weight LLMs beyond APIs.
• RL or preference optimization for LLMs with reward design.
• Strong Python skills, PyTorch or JAX/Flax.
• GPU-based training experiments development and debugging.
• Data curation, experimental design, model evaluation skills.

Responsibilities

• Design and run experiments for post-training alignment using supervised fine-tuning, RL, and reward design.
• Develop agentic language models for sequential patient interactions and tool-augmented workflows.
• Explore knowledge distillation from larger models to smaller efficient ones.
• Create curated datasets, synthetic conversations, and patient-simulation environments.
• Evaluate models with reproducible experiments and clinical review.
• Adapt models to Indian primary care communication patterns and languages.
• Deliver checkpoints, code, results, and documentation of findings and next steps.

Skills

ML research

Open-weight LLMs

RL / Reward design

Python

PyTorch

JAX/Flax

GPU training

Data curation

Experimental design

Model evaluation

Research reading

Job description

Sahai | eGovernments Foundation

About Sahai

Sahai is an initiative backed by the eGovernments Foundation, building an AI triage system and clinician co-pilot to expand primary healthcare access across rural India.

Through voice-based conversations, Sahai gathers patient histories and prepares differential diagnoses and draft care recommendations for qualified doctors to review. We are developing open-weight models suited to Indian primary care, supported by clinical data, safety controls, and rigorous evaluation.

The AI recommends. The doctor decides.

The Role

We are looking for a hands-on AI/ML Scientist / Engineer to develop and validate a proof-of-concept model for clinical interviewing and diagnostic decision support.

The challenge goes beyond teaching models medical facts. It involves improving their ability to conduct useful multi-turn patient interviews, work with clinical tools, and reason with incomplete information.

You will collaborate with clinicians, data specialists, and engineers to design experiments, post-train candidate models, and establish what works. This role focuses on model research and development, not production deployment or operational infrastructure.

What You Will Do

• Post-training and alignment: Design and run experiments using supervised fine-tuning, reinforcement learning, preference optimization, and reward design.
• Agentic model development: Improve language models for sequential patient interactions and tool-augmented clinical workflows.
• Knowledge distillation: Explore transferring capabilities from larger teacher models into smaller, computationally efficient models.
• Training data and simulation: Partner with clinical and data teams to develop curated datasets, synthetic conversations, and patient-simulation environments.
• Rigorous evaluation: Compare models and approaches through reproducible experiments, automated assessments, and clinical review.
• India-specific adaptation: Investigate adaptation to the communication patterns, languages, and resource constraints of Indian primary care.
• Research handover: Deliver model checkpoints, experimental code, evaluation results, and clear documentation of findings, limitations, and recommended next steps.

What You Bring

• A strong research or industry background in machine learning, with hands-on ownership of substantial model-development work.
• Demonstrated experience post-training open-weight LLMs, beyond prompting or integrating model APIs.
• Practical experience with reinforcement learning or preference optimization for LLMs, including reward design, alongside supervised fine-tuning.
• Strong Python skills and proficiency in PyTorch or JAX/Flax.
• Experience running and debugging GPU-based training experiments, with an understanding of compute and memory constraints.
• Sound judgment in data curation, experimental design, and model evaluation.
• The ability to read research critically, implement promising methods, and test whether they work in practice.

We value demonstrated technical depth and experimental rigor more than familiarity with a long list of tools. Prior production-deployment experience is not required.

Especially Valuable

• Experience with knowledge distillation, smaller language models, or model compression.
• Work on multi-turn agents, simulated environments, or multi-agent evaluation.
• Experience with distributed training.
• Work in healthcare AI or another safety-critical domain.
• Experience with multilingual or Indic-language models.
• Relevant publications, open-source contributions, or substantial independent projects.

A medical background is not required. Respect for clinical expertise and responsible handling of sensitive data are essential. Speech-model training is not the primary focus of this role.

What Success Looks Like

An evaluated proof of concept that demonstrates measurable improvements over a baseline, supported by reproducible experiments and an honest account of its limitations. The work should help Sahai decide which approaches merit further development; production readiness is not the expected outcome.

Why Sahai

Apply advanced AI research to a concrete public-health challenge. Help establish the model capabilities that could bring qualified medical judgment within reach of underserved communities.