TrialSync Agent

TrialSync Agent: AI-powered clinical trial site selection in 48 hours, not 6 months. Saves $5M+ per trial using Gemini 3 + MongoDB to match patients, score sites, and automate feasibility.

Comment

Inspiration

The $5 Million Problem Nobody Talks About

As a former database administrator working in clinical trial systems, I watched pharmaceutical companies burn millions of dollars and waste months on a process that should take days: selecting the right clinical trial sites.

The stats are brutal:

  • 6 months average timeline for site identification and feasibility
  • 40% of sites enroll ZERO patients (complete waste of resources)
  • $2M-$10M per trial spent on site selection and activation
  • PhD-level teams manually reading 200-page protocols and querying hundreds of sites

Meanwhile, I saw firsthand how MongoDB's flexible document model could handle the complex, variable structure of clinical trial data. I experienced the frustration of rigid relational schemas trying to accommodate protocols with 12 inclusion criteria versus protocols with 25.

Then came Gemini 3 and Google Cloud Agent Builder.

Suddenly, the pieces clicked: What if an autonomous AI agent could parse complex medical protocols, query patient populations across hundreds of sites, score them on multiple factors, and deliver executive-ready reports—all in 48 hours instead of 6 months?

That's when TrialSync Agent was born. Not as a chatbot. Not as a simple query tool. But as a true autonomous agent that accomplishes a complex multi-step task with real enterprise value.

What it does

TrialSync Agent is an autonomous AI system that revolutionizes clinical trial site selection through 6 core capabilities:

1️⃣ Intelligent Protocol Parsing

Upload a 150-page trial protocol PDF. TrialSync Agent (powered by Gemini 3 Pro) extracts:

  • Inclusion/exclusion criteria
  • Disease codes (ICD-10)
  • Biomarker thresholds (e.g., "NT-proBNP >600 pg/mL")
  • Age ranges, gender requirements, comorbidity restrictions
  • Target enrollment numbers

All structured into MongoDB documents for downstream processing.

2️⃣ Patient Population Matching

The agent queries MongoDB Atlas across 250+ clinical trial sites, searching de-identified patient counts that match protocol criteria. MongoDB's aggregation pipelines handle complex multi-collection joins:

  • Match sites by therapeutic area (e.g., Cardiology)
  • Filter patient populations by ICD-10 codes
  • Apply age, gender, and biomarker filters
  • Calculate matching patient counts per site

3️⃣ Multi-Factor Site Scoring

TrialSync scores every site on 6 weighted factors:

  • Patient Availability (30%): Matching patient count
  • Enrollment Velocity (25%): Historical speed (patients/month)
  • Activation Speed (15%): Time from contract to first patient
  • Retention Rate (15%): % of patients completing trials
  • Diversity Score (10%): Race/ethnicity representation (FDA priority)
  • Cost Efficiency (5%): Per-patient enrollment cost

MongoDB aggregation pipelines compute all scores in a single query, then write results to the site_scores collection.

4️⃣ Automated Feasibility Management

The agent generates customized feasibility surveys for the top 20-30 sites and orchestrates email outreach. MongoDB Change Streams monitor responses in real-time, triggering follow-up reminders every 48 hours.

5️⃣ Executive Reporting

Output: A comprehensive PDF report with:

  • Ranked site list (Top 10-15 with detailed scorecards)
  • Risk analysis (flags for low diversity, slow activation, high cost)
  • Timeline projections (estimated enrollment completion dates)
  • Budget optimization scenarios

6️⃣ Continuous Intelligence

The agent maintains a complete audit trail in MongoDB (FDA 21 CFR Part 11 compliance) and sends real-time alerts via Slack/email when sites respond or risks emerge.

Bottom line: What takes human teams 6 months now takes TrialSync Agent 48 hours, saving $5M+ per trial.

How we built it

Technology Stack

Core Platform:

  • Google Cloud Agent Builder - Orchestrates the multi-step autonomous workflow
  • Gemini 3 Pro - Complex reasoning for protocol parsing and risk analysis
  • Gemini 3 Flash - Fast queries for site scoring and data extraction
  • MongoDB Atlas - Data layer with MCP (Model Context Protocol) integration

MongoDB Architecture (6 Collections):

trialsync/
├── protocols          // Structured protocol data + vector embeddings
├── sites              // Master site data (250+ sites with geospatial indexing)
├── patient_populations // De-identified patient counts by condition
├── site_scores        // Agent-generated rankings and risk flags
├── feasibility_surveys // Outreach tracking with response monitoring
└── agent_logs         // FDA-compliant audit trail (time-series collection)

Built With

  • and
  • and-bespoke-vanilla-css-on-node.js/express
  • change-streams
  • frontend-&-ai:-react
  • geospatial-indexes
  • orchestrated-by-google-cloud-agent-builder-and-powered-by-gemini-3.-data-&-partner-track:-mongodb-atlas-(with-mcp-server)-utilizing-14-stage-aggregations
  • time-series
  • vector-search
  • vite
Share this project:

Updates

Submission history