Platform

Observe

See how your AI handles emotion in production. Observe provides visibility into emotional behavior across your interactions — drift detection, escalation coverage, and intensity patterns that traditional monitoring misses.

The Visibility Problem

AI systems handle emotionally consequential interactions every day. But traditional monitoring tools track errors, latency, and token counts — not whether emotional handling changed after a model upgrade, or whether escalation rules are firing correctly.

The question Observe answers: Did emotional handling drift after your last deployment? Are guardrails overfiring or underfiring? Which high-intensity moments went unhandled?

Input Modes

DeepaData accepts content in different structures depending on your integration pattern. The input mode determines what gets captured and how observation triggers fire.

Topic-shift triggered

The /v1/observe endpoint captures salience at topic-shift boundaries. Lightweight "what changed + why it matters" records without full artifact extraction.

Best for: High-volume chat, continuous monitoring, escalation detection

Passage/session-based

The /v1/extract endpoint processes complete passages or session transcripts into full 96-field EDM artifacts.

Best for: Therapy sessions, long-form journaling, retrospective analysis

Batch processing

The /v1/batch/upload endpoint processes historical content retroactively. Upload JSON or CSV with up to 1,000 records.

Best for: Historical data governance, migration, compliance backfill

Observation Triggers

The Observe endpoint detects 7 trigger types that indicate when emotional context has shifted or requires attention.

TriggerTypeDescription
topic_shiftcontentConversation topic changed meaningfully
affect_shiftcontentEmotional intensity or valence changed significantly
risk_signalcontentContent indicates potential harm or distress
recurrencecontentTheme or pattern has appeared before
user_markcallerCaller explicitly requested capture (force: true)
session_closecallerSession ending, capture final state
time_gapcallerSignificant time elapsed since last observation

Content-detected triggers are identified by the LLM. Caller-provided triggers are passed via request parameters (is_session_close, time_since_last_ms, etc.).

Using the Observe API

curl -X POST https://www.deepadata.com/api/v1/observe \
  -H "Authorization: Bearer dda_live_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "content": "I have been feeling much better about work lately...",
    "subject_id": "user-123",
    "session_id": "session-456",
    "is_session_close": false
  }'

Dashboard Metrics

The Observe dashboard in the platform console surfaces five key metrics from your extraction and observation data.

Emotion Distribution

Primary emotions across extractions

Trigger Distribution

What triggers salience capture

Escalation Coverage

% flagged for escalation

Intensity Histogram

Emotional weight distribution

Observe Products

Beyond the overview dashboard, Observe includes specialized analysis tools.

Drift Detection

Compare emotion intensity distributions across time windows. Detect silent drift after model upgrades or configuration changes.

Escalation Matrix

Correlation view of intensity score vs escalation action. Identify gaps (high emotion, no response) and noise (low emotion, over-response).

Coverage Report

Guardrail coverage analysis. % of high-intensity passages that triggered escalation vs false positives.

Emotional Exposure

How many emotionally consequential interactions this period, by domain and jurisdiction.

Related