Lead-Call Agentic Analysis (Step 1)

This step makes lead web and phone-to-phone calls select a mandatory analysis schema, and enriches every real (non-test) call's analysis with internal agentic blocks used for automation, dashboards, and insights.

What a real call's analysis now stores on Contact

Field	Source	Shown to user?
analysis	user-defined schema features	Yes
conversation_summary	combined LLM call	Yes
quality_evaluation	combined LLM call	Yes (quality)
agenticSignals	combined LLM call	No (internal)
industrySignals	combined LLM call, from schema's industry	No (internal)
conversationFacts	combined LLM call, dynamic key/value	No (internal)
embedding_narrative	combined LLM call	No (internal)
embedding	text-embedding-3-small on the narrative	No (internal)

All produced in one combined LLM call (no extra call) + one cheap embedding call.

agenticSignals (industry-independent, drives the future cron)

call_outcome, answered_but_no_audio, reachability, next_action, callback_requested, callback_time_raw, callback_at (ISO-8601 with +05:30 IST offset), callback_window, preferred_channel, follow_up_priority, user_sentiment, engagement_score, interest_level, contact_restriction, complaint_flag.

industrySignals (per AnalysisSchema.industry)

Chartable dimensions per industry: real_estate, healthcare, education, hospitality, travel, legal, generic. Defined in Analysis/src/pipeline/internal_analysis.py.

conversationFacts (fully dynamic)

Array of { key, value, type, confidence }. The model chooses keys and values, capturing everything actually discussed only-discussed, nothing invented.

Test calls

Test calls extract only analysis (user features). No quality/summary/ agentic/industry/facts/embedding saves tokens. Authority: the request test_call flag and Contact.isTestCall (whichever is true).

UI

On the lead Call tab, Web and Phone-to-phone now require an analysis schema (dropdown, mandatory). The last selection is remembered in localStorage (oktivo:lastLeadCallSchema). Industry is taken from the selected schema no separate industry picker. AI-agent mode keeps deriving the schema from the agent.

---

Backfill: re-analyze existing contacts

Re-runs analysis for contacts that already completed analysis under the old pipeline, to populate the new blocks. Non-destructive (re-trigger only). Test calls are always skipped.

HTTP (curl)

Dry run (default reports what WOULD be re-analyzed, makes no calls):

curl -X POST "$BASE_URL/internal/scripts/reanalyze-contacts" \
  -H "Content-Type: application/json" \
  -H "x-api-secret: $SUPERADMIN_API_SECRET" \
  -d '{ "dryRun": true }'

Real run (actually re-triggers analysis):

curl -X POST "$BASE_URL/internal/scripts/reanalyze-contacts" \
  -H "Content-Type: application/json" \
  -H "x-api-secret: $SUPERADMIN_API_SECRET" \
  -d '{ "dryRun": false }'

Scoped / capped examples:

# Only a specific analysis schema, first 100 contacts
curl -X POST "$BASE_URL/internal/scripts/reanalyze-contacts" \
  -H "Content-Type: application/json" \
  -H "x-api-secret: $SUPERADMIN_API_SECRET" \
  -d '{ "dryRun": false, "analysisSchemaId": "anl_xxx", "limit": 100 }'

# One customer, higher concurrency, include even already-backfilled contacts
curl -X POST "$BASE_URL/internal/scripts/reanalyze-contacts" \
  -H "Content-Type: application/json" \
  -H "x-api-secret: $SUPERADMIN_API_SECRET" \
  -d '{ "dryRun": false, "customerUID": "cust_xxx", "concurrency": 8, "onlyMissingBlocks": false }'

Body options

Field	Default	Meaning
dryRun	true	true = report only, no calls. false = re-trigger.
onlyMissingBlocks	true	Only contacts without agenticSignals (avoid re-paying).
analysisSchemaId		Scope to one schema's contacts.
customerUID		Scope to one customer.
campaignUID		Scope to one campaign.
limit	all	Cap number of contacts.
concurrency	5	Parallel re-trigger requests (1–20).

Standalone CLI (from backend/client_backend)

npx tsx src/scripts/reanalyze-contacts.ts --dry=true
npx tsx src/scripts/reanalyze-contacts.ts --dry=false --limit=100
npx tsx src/scripts/reanalyze-contacts.ts --dry=false --analysisSchemaId=anl_xxx
npx tsx src/scripts/reanalyze-contacts.ts --dry=false --customerUID=cust_xxx --concurrency=8

Response shape

{
  "success": true,
  "dryRun": true,
  "matched": 1234,
  "skippedTestCalls": 20,
  "skippedNoData": 5,
  "triggered": 0,
  "failed": 0,
  "byCallType": { "ai_agent": 1000, "human": 150, "phone_to_phone": 84 },
  "sampleContactUIDs": ["api_...", "crm_..."],
  "errors": []
}