Luna Labs builds measurement instruments for domains that resist quantification, combining machine learning, quantitative finance, and actuarial modelling to replace guesswork with auditable data. The practice develops systems such as a wrist-IMU platform that reconstructs shift activity from inertial streams, a native quant terminal with live/backtest parity, and health-workforce demand models. Every output ships with its uncertainty rendered, and low-confidence results are allowed to abstain rather than guess.
Funding
Funding not disclosed
Founders
Product
Problem
Most consequential work leaves no trace—a shift is walked, a market is read, a risk is carried—and the record afterwards is a guess with a confident tone. Organizations in health services, trading, and operations lack instruments that can measure what actually happened, forcing decisions on unreliable reconstructions and unverified assumptions.
Solution
Luna Labs builds measurement instruments for domains that resist measurement, combining machine learning, quantitative finance, and actuarial studies held simultaneously. The practice treats every quantity as a distribution and every model as something that must survive an audit, with out-of-sample honesty as a core discipline. Systems are built around a fixed reference—a dock, a baseline, a known origin—that makes everything downstream checkable. The approach separates deterministic computation from learned components, uses language models only for language, and renders uncertainty visibly in every interface. Negative results are published with the same typeface as positive ones, so what failed compounds into future system design.
Target Audience
Primary customers are organizations in health economics, workforce planning, quantitative trading, and operational domains that need auditable measurement of work that currently leaves no trace—including health services, specialist practices, and trading operations.
Features
- Wrist-IMU intelligence platform (Ichne) that reconstructs an eight-hour shift's activity timeline, trajectory, and floor map from a single inertial stream and a fixed dock
- Native quant terminal (Chronos) with a Rust application shell, Python and R sidecars, locally hosted models, and one strategy implementation that runs identically in backtest and production
- Factor-graph state estimation and sensor fusion with calibrated confidence and abstention as first-class outputs
- Data pipelines resilient to upstream reshaping, with entity resolution and geospatial joins at scale
- Applied LLM systems using retrieval pipelines, Model Context Protocol tool integration, and planning modes for governed document generation
- Actuarial demand models projecting specialist supply from population age structure, mortality-derived morbidity proxies, and income elasticity
- Scale-to-zero cloud architecture priced for idle, with queue-and-worker designs for long jobs