Health Coach Bot (Marily Oppezzo & Michele Patel)

Field	Value
Status	🟢 Active — Building Prompt Lab + content updates
Category	Build-a-Bot
Contact	Marily Oppezzo (moppezzo@stanford.edu), Michele Lanpher Patel (Michele.patel@stanford.edu)
What	Build-a-Bot spin-off for healthcare context - people building their own AI health coaches
Prototype	coach.bot101.app
Tags	[SPARK] [CO-CREATE] [BRIDGE-BUILD]

Context

Both are Instructors at Stanford Prevention Research Center. Prototype exists. Focus is behavioral health coaching (not prescription) — “exercise snacks” / “Snacktivity” approach for incorporating activity into the workday.

Latest (Mar 24, 2026)

Regroup with Priya Talreja (new team member), Marily, Michele. Major decisions on data architecture and experiment design:

• Cloud syncing = default for IRB (local-only as opt-out). Steer IRB this direction.
• Bot version tracking — add metadata to conversation logs so each message records which bot configuration was active. Essential for research analysis.
• A/B experiment proposed — editable vs non-editable bots. “Coaching your coach” narrative. Measure edit counts vs behavior change.
• Moderation pipeline — keyword filter → OpenAI moderation → custom prompt (7 categories). Adds latency but necessary.
• Provenance — health responses need peer-reviewed source citations, not commercial links.
• Firebase concerns — explore medical school storage alternatives for health-adjacent data.
• IRB next steps — meet with Shelley (out until Mar 27) to review required features. Prepare consent materials.
• Follow-up meeting being scheduled (possible Tuesday). Caitlin may join.

Previous (Mar 3): Prompt Lab, BCT content, “About Me” clarity, RAG links, data export — still open.

Bug: Save persistence issue in health chat still needs investigation.

Detailed notes: See ../meetings/running-notes.md (Mar 24 entry), ../people/marily-oppezzo.md, ../people/michele-patel.md