Accelerating Personalized Health Protocols with Windmill's Agentic AI Pipeline
Solution for a precision health company offering longevity protocols based on biomarkers.
Project details and story
- Industry
- Healthcare
- Solution
- Agentic AI Discovery Sprint
- Agentic AI Architecture
- Proof-of-Concept (PoC) Build
- AI Agent Engineering
- DevSecOps & Deployment
- Integration Engineering
- Human-Centered UX / UI Design
- Year
- 2025
Challenges
Manual generation of personalized health protocols was slow (~2 hours), inconsistent, and resource-intensive. This hindered scalability and user experience.
Approach
We partnered with xLongevity to streamline protocol generation by building a modular agentic AI pipeline. Specialized agents were designed to extract data, assess risk, and compose personalized plans—delivering consistent, safe outputs in under five minutes.
Solution
Windmill's modular agentic AI pipeline automated:
- Parsing of clinical documents (PDFs, notes)
- Mapping of biomarkers, PhenoAge, NAD+, senescence risk
- Supplement and intervention matching
- Personalized dietary and lifestyle planning
- Summary generation in structured JSON for seamless UI/report rendering
Architecture & Technology Stack
Frontend (React Forms)
Clinicians upload reports or input data; AI suggestions appear inline.
Tech Stack: React, JSON Schema Renderer
Agentic AI Orchestrator
Manages specialized agents for parsing, risk analysis, protocol generation.
Tech Stack: LangChain, Azure OpenAI (GPT-4, GPT-4o)
Backend APIs
Stores structured data, exposes APIs, enables exports and audit tracking.
Tech Stack: Java, Django, PostgreSQL
Agentic AI Pipeline Design
Intake Parser Agent
Extracts structured health data from text/PDFs
Model: GPT-4o + Textract
Output: JSON metadata
Risk Assessor Agent
Flags contraindications, red flags, conflicts
Model: LLM + internal logic
Output: Risk report
Intervention Matcher Agent
Suggests clinical-grade therapies (e.g., NAD+, etc.)
Model: GPT-4
Output: JSON protocol section
Supplements Matcher Agent
Matches supplements with safety checks
Model: GPT-4o + Textract
Output: Supplement list
Protocol Composer Agent
Merges outputs into safe, sequenced plan
Model: GPT-4
Output: Final protocol JSON
Diet Recommendations Agent
Personalized dietary strategy
Model: GPT-4
Output: JSON meal guide
Lifestyle Recommendations Agent
Suggests habits for sleep, stress, exercise
Model: GPT-4
Output: Lifestyle suggestions
Summary & Report Generator
Final clinical summary in JSON + HTML
Model: GPT-4o
Output: Report document
Impact
Before
Manual health protocol workflows
~2 Hours
Manual protocol generation
High
Clinician review required at every step
Variable
Outcomes differed across clinicians
After
AI–powered protocol automation
<5 Minutes
AI composes protocols in minutes
Minimal
Light validation only when needed
Consistent
AI ensures repeatable, reliable results
