BRAT Research Platform

Product Lead & AI Engineer

Research platform

Built conversational practice, study onboarding, and researcher review tools for a UT Austin parenting study.

Overview

BRAT (Boundaries, Resilience and Affection Training) is a conversational role-play and coaching platform where parents practice difficult conversations with a simulated child before trying them at home. It operationalizes PMTO (Parent Management Training, Oregon), an evidence-based parent-training model, into LLM-driven role-play with technique-level feedback. BRAT is a production research platform for a UT Austin parenting study, developed with collaborators across partner universities.

My contribution

As product lead and AI engineer, I translated research requirements into the platform's design and implementation. I built the conversational system, the safety and moderation layer, and the researcher-facing tooling, and integrated the study-operations workflows that move participants from screening through practice and review.

Platform & Conversational System

  • Built the role-play and coaching platform on Next.js, React, and Node, deployed on Azure App Service with Azure Blob Storage and a PostgreSQL data layer.
  • Implemented PMTO-informed conversational fidelity metrics across dialogue turns, giving researchers a structured way to assess session quality.
  • Contributed safety guardrails, input and output moderation rules, and session-review workflows for participant-facing interactions.

Study Operations & Researcher Tooling

  • Integrated Qualtrics-based screening, consent, and onboarding into the platform, standardizing participant handoff.
  • Built a researcher dashboard for session retrieval, transcript review, and participant progress monitoring.
  • Built a quantitative data pipeline enabling clinician and researcher review of multilingual interaction data.

Technologies & Tools

Next.jsReactNode.jsAzure App ServiceAzure Blob StoragePostgreSQLQualtricsMulti-agent LLM design

Deliverables & outcomes

  • Deployed a production research platform supporting a UT Austin parenting study.
  • Implemented PMTO-informed fidelity metrics that let researchers assess session quality at the level of technique.
  • Standardized participant screening, consent, and onboarding through integrated Qualtrics workflows.
  • Built researcher tooling for transcript review and progress monitoring across multilingual sessions.

Links & Resources