Selected work / Seed Health
From working demo to integrated member experience
I built a personalization and learn-to-earn journey that moved from a working demo through a measured pilot into Seed’s broader member product.
The arc
A demo built to become a journey.
The outcome was more than an AI assistant. It was a connected product experience for members.
Working product flow
Onboarding, product-fit questions, personalized results, learning modules, learn-to-earn, and Agent Flora existed together in the initial demo.
Measured engagement
A six-week pilot reached 40 weekly active users and showed a 24-point increase in module completion.
Broader member experience
The personalization and learn-to-earn journey became part of Seed’s wider member product. Agent Flora’s later status is unknown.
Member journey
From first question to the next useful step.
The flow linked what a member shared to relevant education and product guidance.
Give the experience enough context to begin.
Capture structured answers and open-ended needs.
Turn responses into tailored guidance.
Connect recommendations with education and progress.
Reward continued learning inside the member experience.
Related AI systems
Two users. Two different jobs.
Both systems depended on trustworthy source retrieval, but they served different workflows.
Agent Flora
A retrieval-backed assistant inside the working member demo, connected to the broader quiz and module journey. Its status after the demo is not something I can verify.
Flask · OpenAI embeddings · Pinecone · grounded response generation
Article Generator
An internal Streamlit tool that retrieved indexed source material before drafting content for human review. The workflow included document preparation, chunking, indexing, and citation-focused evaluation.
Streamlit · LangChain · Pinecone · reviewed drafting
Internal article generator
From source material to a reviewable first draft.
This was a separate internal tool with its own workflow and measured outcomes, not an extra screen in the member demo.
The bottleneck was getting from a scattered research and content library to a draft a subject-matter expert could evaluate. I built a Streamlit workflow that made the source trail visible before publication decisions.
- 01 / PrepareMake sources usable
Clean and segment internal content, SciCare Q&A, and relevant PubMed research for retrieval rather than handing a model one large document.
- 02 / RetrieveFind the evidence first
Index source chunks with embeddings and retrieve material relevant to the draft brief. This made grounding an explicit input to generation.
- 03 / DraftGenerate an editable starting point
Use the retrieved context to assemble a first draft with citations that a human editor could inspect and revise.
- 04 / ReviewKeep the expert in the loop
Route the output to subject-matter review. The workflow supported editorial judgment rather than treating a generated article as publishable by default.
The measured first-draft time reached 2.5 minutes; citation coverage increased from 72% to 88%. Those figures describe this internal drafting workflow, separate from the member pilot.
Measured outcomes
Results without collapsing the stories.
The pilot measured member engagement. The internal tool had its own drafting and citation outcomes.
weekly active users
module-completion increase
time to first draft
citation coverage
Engineering decisions
Grounded AI is a product system.
The member demo combined session-backed onboarding, questionnaire state, recommendation logic, module progression, and retrieval-backed answers. In the internal tool, preparing documents for retrieval and preserving the review step were as important as the generation call itself.
The production code path after integration is proprietary. This case study focuses on the experience I built, the pilot we measured, and the product direction you can accurately trace from that work.
Next project
QuietDesk
A quieter Mac desktop, built around one precise interaction.
Explore the case study →Build note
When a demo becomes a workflow
The product lesson behind the Seed experience.
Read the note →