An NSF-backed generative AI platform that translates dense regulatory documents from CMS, FDA, and ONC into persona-specific explanations, gathers structured public comment, and submits it back into federal feedback systems. Built to close the gap between who healthcare policy affects and who actually shapes it.
Trust and transparency over accuracy. Co-creation over deployment. Emotional intelligence and systems design over model outputs.
Healthcare policy is written in language that excludes the people it governs. Feedback systems reward the actors who already have power. The public-comment infrastructure is a pipeline designed for institutions, not for patients.
Regulatory language is a barrier that deters community engagement before it starts. The people most affected by policy changes are the least equipped to respond to them in the window feedback is open.
Rural voices, patients, and marginalized communities are systemically excluded from public comment periods — not because they have nothing to say, but because the system assumes legal fluency they were never asked to acquire.
Even when comments are submitted, the way they're analyzed doesn't turn them into actionable policy signal. Regulators receive noise; communities get no response.
An AI layer between federal policy documents and the public — designed to reduce cognitive load on the citizen side and improve signal quality on the regulator side.
GPT-4 with a custom RAG architecture summarizes policies, translates jargon, and generates glossaries. Users can ask questions in natural language and receive persona-appropriate answers with source citations.
Guided flows tailored to patients, advocates, hospital staff, and economists — each surfacing the parts of a policy document most relevant to their context, in the language that context uses.
Collects, clusters, and formats user submissions into structured packets for CMS and ONC public comment portals. Turns individual comment into thematic signal regulators can act on.
25+ NSF iCorps customer discovery interviews. Stakeholder interviews across hospital admins, rural health advocates, health economists, and policy researchers. Journey mapping and pain point identification.
Persona-based flow design, AI prompt engineering, data schema mapping. Early rapid prototypes built to test whether persona-specific explanation of a single policy actually changed comprehension.
RAG architecture for document parsing, keyword extraction, and semantic clustering. Custom prompts for persona-specific Q&A and comment generation. Pipeline engineered for source-traceability and transparency.
Beta testing with hospital staff and policy researchers. CMS/ONC integration for real-time policy comment submission. Dashboard built to visualize feedback patterns and coverage gaps for regulators.
Before and after, across the four outcomes we set out to move: comprehension, accessibility, response time, and feedback quality.
The platform was designed for three primary personas identified through NSF iCorps discovery. Each sees a different view of the same document.
Surfaces the fiscal implications of proposed policy — what it does to reimbursement, coverage, and regional health budgets over time.
Aggregates rural-specific input and translates it into submission-ready language for comment portals — so rural voices arrive at CMS as signal, not noise.
Isolates implementation implications of policy changes — what it costs hospitals to comply, where the funding gaps sit, and what to ask regulators to adjust.
Prompt templates tuned per persona and per use case. Output formatted for downstream comment-portal submission, not for conversational output.
Pre-processing layer that structures dense regulatory text into retrievable chunks, surfaced to the LLM with source traceability preserved throughout.
A regulator-facing view that turns individual submissions into thematic clusters, highlighting where comment volume is concentrated and where it's missing.
Every submission is auditable. Users see exactly what was generated, what was edited, and what was sent — preserving consent and public record integrity.
In civic AI, a model that is 95% accurate but opaque performs worse than one that is 85% accurate and shows its reasoning. We over-invested in data stewardship and user control — which turned out to be the feature that actually drove adoption. Accuracy without provenance is just another black box.
Diverse stakeholder input shaped everything from interface language to use case prioritization. Economists, advocates, hospital administrators, and patients used different vocabularies for the same concept — and getting that vocabulary right was the difference between a tool they used and a tool they abandoned.
Success hinged on emotional intelligence and systems design, not model outputs. The LLM was one component inside a larger choreography of trust, context, and control. Teams that reverse that ratio — model-first, experience-second — build demos, not products.
"Christine's standout quality is her constant curiosity, which she demonstrated in recent projects leveraging generative AI to drive innovation. Her contributions to projects involving CapsicoHealth's AI-driven analytics products helped demonstrate the value of these tools to stakeholders. Her ability to articulate the business and clinical implications of cutting-edge technologies makes her an invaluable collaborator in aligning AI capabilities with healthcare objectives."Dr. J. Sairamesh CEO · CapsicoHealth
Open to senior strategy and applied AI roles at the intersection of healthcare, civic infrastructure, and public interest technology — particularly where research rigor, behavioral design, and engineering judgment all have to sit in the same person.
Start a conversation →