About LUNA:Luna is a health intelligence company. We build AI-native wearables (Luna Ring and Luna Band) and LifeOS, the platform that turns their data into daily guidance.At the centre of LifeOS is a personal health agent that knows your health data, talks with you, and acts on your behalf.About the role:You will build the data science behind the Luna health agent. That means two things: making data readable enough for an agent to act on, and proving with real outcomes that the agent's guidance works.We use frontier language models and build what sits around them: the memory and context layer, and a closed-loop evaluation dataset drawn from what our sensors measure after each recommendation. This role sits at the centre of both.What you will do:- Build the signal layer. Turn raw heart rate, HRV, sleep, and activity data into clean per-user baselines and features the agent can reason on, handling noise, gaps, and device drift.- Design the evaluation system. Create eval sets, automated graders, and regression checks for every agent behaviour, before and after release.- Close the loop. Link each recommendation to what the sensors show afterwards, and turn that into a dataset that makes the agent better over time.- Shape memory and context. Work with engineering on what the agent stores about a user, what it retrieves, and what goes into each prompt.- Run experiments. Use holdouts, A/B tests, and n-of-1 designs on small cohorts to separate real impact from natural improvement.- Model adherence. Learn which nudge, timing, and channel gets a given person to act, since acting on the data is the outcome we care about.- Own cost and quality trade-offs. Track token cost, caching, and latency per user,
and keep quality high as usage grows.- Validate with science. Work with our scientific advisors on validation studies and on the safe limits of health guidance.What you bring:- 4 to 7 years in applied data science or machine learning, with ownership of work that reached production.- At least one shipped LLM-based product feature, including the evaluation you built for it.- Strong statistics: causal inference, experiment design, and comfort drawing conclusions from small samples.- Fluent Python and SQL, and a habit of shipping alongside backend engineers.- Sound judgement in health: you know where guidance ends and medical advice begins.- Clear writing. You can explain a result to a product team and to a scientist.Technical skills:Must have:- Python: pandas, NumPy, scikit-learn, and SciPy or statsmodels for statistics.- SQL: window functions, joins on event data, and query optimisation on large tables.- LLM APIs: prompt design, structured outputs, tool calling, and prompt caching.- LLM evaluation: eval sets, LLM-as-judge graders, and an evaluation framework or a custom harness.- Statistics: hypothesis testing, confidence intervals, power analysis, regression, and mixed-effects models.- Experimentation: A/B test design, holdouts, sequential testing, and difference-in-differences.- Git and code review:
code that others can run and maintain.Strongly preferred:- Retrieval and memory: embeddings, vector search, hybrid search, and reranking.- Data stores: Postgres, plus a warehouse such as BigQuery or Snowflake.- Pipelines: dbt, Airflow, or a similar orchestrator.- Causal inference: propensity scores, synthetic controls, and libraries such as DoWhy or CausalImpact.- Product analytics: Mixpanel or Amplitude, including event taxonomy design.- Cloud basics: AWS or GCP, enough to deploy and monitor your own work.Bonus:- Time-series methods: resampling, smoothing, anomaly detection, and changepoint detection.- Signal processing: filtering and feature extraction from heart rate, HRV, and accelerometer data.- ML modelling: gradient boosting (XGBoost, LightGBM) and PyTorch basics.- Recommenders and bandits: contextual bandits for nudge timing and selection.- Agent frameworks: tool-using agents, plus tracing or observability tools.- R: useful for statistics and research collaboration, not required if your Python statistics are strong.Positive to have:- Hands-on experience with physiological or wearable time-series data (heart rate, HRV, sleep, activity).- Retrieval, memory systems, or context engineering for agents.- Voice or conversational product experience.- Work with clinical data such as blood reports or medical records.- Recommender systems or behaviour change modelling.- Published research or validation studies in digital health.Who this role is not for:- You are looking for a reporting and dashboards role.- You prefer a settled roadmap. This is early work with a small team and a lot of open questions. (ref:hirist.tech)
📌 Luna - Applied Data Scientist - LLM/HealthTech (Gurugram)
🏢 Luna
📍 Gurugram