How It Works
New to Project Ignaz?
Take a guided walkthrough of the major pages and workflows.
How It Works
Project Ignaz streamlines Event-Based Surveillance (EBS) through four simple stages — from collecting data across multiple sources to delivering verified, analyst-ready reports. Click each step to learn more.
Step 1
Data Collection
The system automatically gathers disease-related information from three major data sources every night, then combines them into a single unified view. Analysts can also add sources manually.
1.1 — Nightly Automated Collection
Every night, automated jobs run to search for new disease outbreak information across the web and specialized surveillance databases. Each pathogen-location pair configured in the system gets checked for the latest news and reports.
1.2 — Data Sources
Project Ignaz pulls from three complementary data sources to provide broad and deep coverage of global public health events:
Automated web searches surface the latest news articles, official health authority reports, and publications about disease outbreaks worldwide. Each search is tailored to a specific pathogen and location.
The WHO's Epidemic Intelligence from Open Sources platform provides curated, real-time intelligence on public health events detected from open sources around the world.
The Global Infectious Diseases and Epidemiology Network provides comprehensive outbreak data, disease profiles, and historical epidemiology across countries and pathogens.
1.3 — Consolidation
Results from all three sources are combined into a single unified database. Duplicate articles are automatically detected and filtered out, and each source is tagged with where it came from for traceability.
Duplicate articles are automatically detected and removed
Each source is tagged (Serper, EIOS, or GIDEON) for traceability
New sources are automatically queued for AI analysis
1.4 — Manual Source Entry
In addition to automated collection, analysts can manually submit sources through the Signal Management page. Three input modes are supported:
Paste an article URL. The system captures the link and allows the analyst to add a title and notes.
Free-form text input for intelligence from offline sources, phone calls, or field reports.
Upload documents (PDF, DOCX, etc.) for signals from reports or publications not available online.
Manually submitted sources go through the same AI analysis as automated ones, ensuring consistent processing.
Step 2
Signal Identification
Once sources are collected, AI continuously scans them to extract key information — identifying the disease, location, severity, and generating a summary. Sources are then scored and ranked so analysts can focus on the most important events first.
2.1 — Continuous Scanning
Location scans are performed directly inside the application through the Location Scanning feature. Analysts can trigger scans for all locations at once or for a single location, and the system analyzes recent signals to generate AI intelligence summaries. New unprocessed sources are automatically picked up and analyzed.
Check Interval
Processed at Once
Auto Retries
Always Running
2.2 — AI Analysis
Each source is analyzed by Azure AI Foundry, which reads the article title, snippet, and (when available) the full linked article to extract structured information. The AI identifies key details about the public health event described in the source.
2.3 — Priority Scoring
After scanning, each pathogen-location pair receives a weighted priority score based on six factors. This score determines the order signals appear in the management interface, guiding analysts to the most urgent situations first.
Mean severity across all sources in the signal
Total number of sources reporting this event
Number of distinct ingestion sources reporting
How recently sources were published or ingested
Highest single severity score in the group
Whether source volume is increasing over time
2.4 — Signal Grouping & Ranking
Scanned sources are automatically grouped by their disease and location. Each group represents a potential public health event. Groups are displayed sorted by priority score — highest priority first — so analysts can efficiently focus on the most critical situations.
Each signal shows the disease name, location, source count, severity, and priority score. Analysts can expand a signal to review individual sources, read AI summaries, and visit source URLs before deciding how to triage.
Step 3
Report Creation
After analysts review the sources, the system assembles all the relevant information and sends it to AI to generate a structured, CDC-formatted Event Report or Watchlist report.
3.1 — Signal Triage
Before a report can be generated, analysts must review and classify each signal in a pathogen-location group. Each signal is marked with one of three verification statuses:
Signal is relevant and will be incorporated into the AI-generated report as source material.
Signal is off-topic, duplicative, or low-quality. Excluded from report generation but retained in the database.
Signal is not relevant to any current public health event. Hidden from management views but preserved.
3.2 — Information Assembly
The system gathers all the relevant context needed to generate a comprehensive report, combining multiple data sources into a single structured package for the AI.
All sources marked "Include" for the disease-location pair — titles, URLs, summaries, severity scores, and publication dates.
Geographic metadata — country, region, WHO classification, population data, and any relevant location-specific context.
Disease-specific information — transmission mode, incubation period, treatment protocols, known epidemiology.
Relevant CDC activity records that provide context on ongoing response efforts, guidance, or travel notices.
Instructions that define the exact report structure the AI must follow, ensuring consistent and reliable output.
3.3 — AI Report Generation
The assembled information is sent to Azure AI Foundry, which generates the full report. During generation, the AI can also read source article URLs directly and perform live web searches to find the most current data:
The AI reads the full content of source article URLs, extracting additional details beyond the initial snippet text.
The AI can perform live searches during generation to find the most current data, statistics, and official statements.
The AI produces a structured report that is automatically parsed into the appropriate report format.
3.4 — Report Types & Structure
The analyst chooses the report type before generation. Each type produces a different output structure tailored to the situation:
Full CDC-style Event-Based Surveillance report with structured fields following the standard EBS methodology:
- At-a-Glance — One-paragraph executive overview
- Public Health Event — Disease/agent classification
- Location & Geoscope — Geographic scope analysis
- Epidemiological Data — Cases, deaths, CFR, date ranges
- Trend Indicators — Increasing / Stable / Decreasing
- Public Health Impact — Severity assessment
- Description — Detailed narrative
- Sources — Attributed URL list with titles
Shorter-form monitoring report for situations that don't yet warrant a full EBS report but require tracking:
- Situational Summary — Brief overview of the event
- Key Metrics — Available case/death counts
- Monitoring Rationale — Why this warrants watching
- Escalation Conditions — Triggers for upgrade to full EBS
- Sources — Attributed URL list
3.5 — Draft Storage
The AI-generated report is saved as a Draft. The report is linked to the disease and location(s) from the signal, and all included source IDs are preserved for traceability.
Not visible on public pages. Requires analyst verification before publishing.
Report retains references to all sources for audit and regeneration.
Source URLs are stored for rendering in report detail views.
Step 4
Verification & Delivery
AI-generated reports require human review before reaching public-facing pages. Analysts verify content accuracy, map geographic coordinates, and mark reports as verified — at which point they are distributed across all presentation surfaces.
4.1 — Draft Review
Draft reports appear in the Manage Reports interface, where analysts can open each report for detailed review. The full AI-generated content is displayed alongside the original source signals for cross-referencing.
AI-generated, pending review
Edit, verify, map coordinates
Published to all pages
4.2 — Content Editing
Every field in the report is editable. Analysts can correct factual errors, update case counts with the latest data, adjust severity levels, refine the narrative, and add contextual notes that the AI may have missed.
4.3 — Geographic Mapping
Analysts pin geographic coordinates onto each report to place events on interactive maps. Coordinates are positioned by analysts using a built-in map editor with location search.
Primary event location with optional radius buffer zone (e.g., city center with 50km range).
Polygon shapes to define containment zones or outbreak areas on the map.
Notable locations within a report area — hospitals, airports, markets, or other relevant sites.
Circle drawn on the map to represent an approximate affected area.
Search for locations by name to find coordinates. The system returns latitude/longitude and geographic boundaries for countries and regions. Click anywhere on the map to place a pin.
All coordinates for a report are saved together as a batch, replacing any previous coordinates. This ensures the map always reflects the analyst's latest placement decisions.
Map display: Verified report coordinates appear on interactive maps. Centroids show as colored markers with tooltips, boundaries as dashed overlays, and area circles as shaded regions. Maps automatically zoom to fit all features.
4.4 — Verification & Publishing
Once the analyst is satisfied with the report content and coordinates, they mark the report as Verified. This instantly makes the report visible on all public-facing pages throughout the application.
4.5 — Presentation Surfaces
Verified reports are automatically distributed across all public-facing pages in the application. Each page provides a different lens into the same underlying data:
High-level overview with AI-generated executive narrative, top threats by severity, and key statistics across all regions.
Domestic events filtered by U.S. states with interactive map, geocoordinate overlays, and severity-colored markers.
Global events organized by WHO region with geographic heatmaps, country-level drill-down, and regional aggregation.
Cross-cutting pathogen view showing all locations affected by a specific disease or agent, with report count and severity summary.
Verified reports also feed into the Newsletter generator, which produces a formatted weekly digest aggregating the most significant events across all regions for email distribution.