AI Engineer

Don’t be data poor — Anuj Iravane, Anterior

Published 2026-08-19 · Duration 16:46

Summary

The talk addresses the critical problem of 'data poverty' in highly regulated domains like healthcare, where the most valuable data (Patient Health Information or PHI) is ephemeral and legally prohibited from being retained, anonymized, or derived for dataset creation. The core solution presented is synthetic data generation. This process involves reversing the standard inference workflow—starting by sampling a desired label and reasoning trace, and then generating the necessary unstructured medical record that would have produced it. The resulting pipeline uses an LLM-based, coarse-to-fine approach, ensuring high fidelity while maintaining domain expert control.

Download summary

Key takeaways

  1. Reverse Inference for Data Generation 5:20

    Instead of running the forward task (Unstructured Data + Policy $\rightarrow$ Label), the method reverses this by sampling a label and a reasoning trace first, then generating the input data that supports it. This circumvents the diversity problem inherent in standard LLM generation.

  2. Domain Expert Ownership (Human-in-the-Loop) 11:30

    To ensure generated data is useful, domain experts (clinicians) must own the pipeline. This is achieved by enabling them to interject at any point in generation and modeling the entire workflow as a skills-based system running on an agent harness.

  3. Synthetic Data Fidelity 14:35

    The generated data can be highly accurate, with early results showing that in a blind review, clinicians were only able to distinguish synthetic from real records about 60% of the time.

Technical details

  • Policy Modeling 380s

    Policies are modeled explicitly as symbolic decision trees (e.g., for a CPAP device medical necessity review). This allows the system to deterministically sample diverse reasoning traces, providing a more uniform and effective prior distribution than standard LLM sampling.

  • Synthetic Pipeline Architecture 520s

    The pipeline uses an LLM-based coarse-to-fine pattern: 1) Generate patient invariants (e.g., sex, birth date). 2) Produce a high-level 'patient journey' (list of events/encounters). 3) Fan out to generate specific documents for each encounter, hydrating them with synthetic information. 4) Include a refinement loop that checks for consistency and contradictions across independently generated documents.

  • Data Handling & Constraints 120s

    The data is highly unstructured (scanned fax bundles, handwriting, tables, images) and classified as PHI. Contracts prohibit retaining, reusing, or deriving information from this source, necessitating synthetic generation.

Mentioned resources

Channel & topics

Watch on YouTube · Back to latest

This independent, AI-assisted summary is provided for commentary and informational purposes. It may contain errors or omit important context. Please watch the original video for the creator's complete presentation. Video, thumbnail, and related copyrights belong to their respective owners.