Explore real workplace analysis scenarios powered by Indicia AI's multi-disciplinary investigation engine. Each case demonstrates how legal, psychological, and HR analyses work together to surface risk patterns, relationship dynamics, and actionable recommendations.
The dashboard below fetches live data from the API — the same data pipeline used in production analysis. Select a case to explore its timeline, participants, risk flags, incidents, and relationship mappings.
Each case study follows a six-layer data pipeline — from raw file upload through to multi-disciplinary AI analysis with mathematical anomaly detection.
Raw communication files are uploaded to a case: email thread exports (.eml), Slack JSON exports, and supporting documents such as performance reviews (PDF). Each source records its type, original filename, raw content, and metadata (platform, message count, export date).
People are identified from the sources. Each participant has a communication role (sender, recipient, cc, mentioned), an organisational role (job title), email address, and aliases used for cross-referencing across different source files.
Individual communications are extracted from the sources — each email in a thread and each Slack message becomes one structured record. Messages are linked to their source, their sender, and ordered by sequence number for chronological reconstruction.
AI models run separate analyses per discipline (legal, psychological, HR). Each analysis produces four structured outputs:
Ten mathematical and information-theoretic techniques are applied to each participant's communication patterns, producing an anomaly score (0–1) and verdict per participant plus an aggregate score for the case.
| Technique | What it measures |
|---|---|
| Cauchy Convergence | Whether communication patterns stabilise or diverge over time |
| Wasserstein Distance | Distributional drift between expected and observed patterns |
| KL-Divergence | Information-theoretic divergence from baseline |
| Fisher Information | Information content and parameter stability of behavioural signals |
| Lyapunov Exponent | Chaotic dynamics and unpredictability |
| Kolmogorov Complexity | Structural complexity and compressibility |
| Mutual Information | Statistical dependency between participants' patterns |
| Persistent Homology | Topological structure in relational data |
| Narrative Entropy | Coherence, topic drift, and contradiction in accounts |
| Minimum Description Length | Model complexity vs explanatory power |
The contrast between participants is the signal: an aggressor typically scores high anomaly across techniques (chaotic, high-drift, highly-divergent), while a target scores low (stable, consistent, coherent).
The dashboard above fetches all six data layers via a single API call and renders seven tabs — each computing its own charts and visualisations client-side from the structured data:
No chart data is hardcoded. All calculations and visualisations are derived dynamically from the API response.
The cases shown above are populated with synthetic seed data — fabricated but realistic workplace scenarios with hand-written messages and hand-crafted analysis outputs. In a production system, the analyses and technique scores would be generated by AI models rather than written by hand. The seed data demonstrates what those models' structured outputs look like once rendered.