AI Insights & CCTV

Retrieves facility check-in / visit history from the Member Management System (MMS) partner integration.

Shows member swipes and visits recorded by door readers, kiosks, apps, or manual entry. Each check-in represents a verified visit event pushed from the MMS.

Data returned per check-in:

• checkInId, facilityId, externalMemberId, timestamp (ISO 8601)
• check_in_type: member, class, staff, guest, trial
• source: how it was captured (manual, tag, rfid, kiosk, app, door_reader, booking)
• class_id: optional, if tied to a class booking

Member enrichment:
Members are returned as a separate array (matched by externalMemberId) with:

• firstName, lastName, gender, dateOfBirth
• membershipStatus, membershipPlans, memberType
• profilePicture, mobilePhone

Filtering:

• Date range via startDate/endDate
• checkInType: filter by visit type (e.g., only "member" check-ins)
• source: filter by capture method (e.g., only "door_reader" swipes)

Use cases:

• Analyze daily visit volume and patterns
• Cross-reference check-in data with AI face detection data (insights_timeline_search)
• Identify members who check in but don't appear on cameras (or vice versa)
• Track class attendance via class-type check-ins
• Verify visit frequency for membership disputes

Correlates unidentified people (from AI vision) with MMS check-in records by timestamp proximity to find likely identity matches.

How it works:
For each unidentified person with multiple visits, the tool looks for check-in swipes that occurred within a configurable lookback window (default 60 minutes) before each detection. If the same member consistently checks in before the same unidentified person is detected, the probability of a match increases.

Use cases:

• Find unidentified people who are actually known members without profile photos
• Identify duplicate unidentified profiles that belong to the same member
• Generate a report of likely matches for manual review before auto-linking
• Audit facility identification coverage

Configurable parameters:

• lookbackMinutes: How far before a detection to search for check-ins (default 60)
• minVisits: Minimum detection sessions required (default 3). More visits = more reliable matching.
• minProbability: Minimum match score to include (default 0.3)

Output includes:

• Ranked matches with probability scores and confidence tiers (high/medium/low)
• Matched timestamp samples showing detection vs check-in times
• Duplicate groups: multiple unidentified profiles matching the same member
• Processing metadata (volumes, timing)

Confidence levels:

• high (80%+, 0-1 other candidates): strong match, candidate for auto-linking
• medium (50-79%): worth manual review
• low (30-49%): needs more data, informational only

Important: This is a read-only analysis tool. It does not modify any records or link profiles.

Part of the AI Insights & People module.

Searches people registered in a facility's AI Insights system by name, membership status, or other attributes.

Supporting fast full-text search with fuzzy matching. Returns person profiles with membership data and faceted counts.

Use cases:

• Find a person by name: query "John Smith"
• List all people with expired memberships: membershipStatus "expired"
• Browse all identified people sorted by name: sortBy "lastName", sortOrder "asc"

Data returned per person:

• Identity: personID, firstName, lastName, fullName, gender, dateOfBirth
• Membership: membershipStatus (active/expired/on_hold/none), memberType (member/guest/staff), externalMemberId
• Profile: profilePicture URL, isIdentified, faceValidationPassed, created, updated
• Attention flags: needsReview, reviewReason, attentionReasons (array)
• Tags: custom tags assigned to the person

Facets returned:

• membershipStatus: count by status (active, expired, on_hold, etc.)
• memberType: count by type (member, guest, staff, etc.)

Use this to find specific people, then drill into their activity with insights_person_get and insights_person_activity.

Part of the AI Insights & People module.

Retrieves the face detection history for a specific person over a date range — their visit log.

Shows every time the person was detected by cameras, with timestamps, which camera saw them, and optionally images and demographics.

Detail levels (use the 'detail' parameter):

• "summary": Per detection: timestamp, camera ID, start/end unix only. Minimal payload for counting visits and calculating patterns.
• "standard" (default): Adds image URLs, similarity score, dominantGender (from multi-frame DeepFace analysis), genderBreakdown, age range. Good for reviewing visit quality.
• "full": Everything including raw emotion data per detection.

Data returned:

• detections: Array of face detection records (capped at 200 per request)
• person: The person's profile (enriched with membership data)
• devices: Camera devices the person was seen on (sensitive fields stripped)
• meta: Query parameters (timezone, date range, unix timestamps)
• nextToken: Pagination token for fetching the next page

Use cases:

• Track visit frequency for a specific member over time
• Analyze which areas/cameras a person visits most
• Check when an expired member was last actively using the facility
• Verify visit patterns for membership disputes

Use after insights_person_get or insights_timeline_search to drill into a specific person's history.

Part of the AI Insights & People module.

Retrieves the full profile of a specific person in the AI Insights system, enriched with membership data.

Returns identity information, membership details, and system metadata for a single person.

Data returned:

• Identity: personID, firstName, lastName, clubID
• Face: avatar/profile picture URL, collectionId, isIdentified, type (identified/unidentified)
• Membership: memberId, memberID (external), membershipStatus, membershipPlans (with start/end dates), memberType, mobilePhone
• Timestamps: created, updated (unix timestamps)

When to use:

• After finding a person via insights_people_search or insights_timeline_search
• To get the full profile before querying their activity history with insights_person_activity
• To check membership details, plans, and contact information

Retrieves facility-wide AI Insights report snapshots for a date range.

This is the best starting point for understanding facility activity. Returns aggregated KPIs and breakdowns for the specified period.

Detail levels (use the 'detail' parameter):

• "summary": KPIs only (~320 bytes) — totalDetections, uniquePeople, gender split, identified/unsure counts, avgDurationSecs
• "standard" (default): KPIs + per-camera counts and age demographics (~3KB)
• "full": Everything including per-camera emotion distributions, hourly detection patterns (24-value arrays), and daily breakdowns (~40-170KB depending on date range)

KPIs returned:

• totalDetections, totalTracks, uniquePeople, uniquePeopleMale, uniquePeopleFemale
• avgDurationSecs (average visit duration in seconds)
• identifiedCount, unsureCount, blacklistCount, unknownCount, checkInCount, expiredCount

Per-camera data (standard/full):

• Detection counts, unique people, age buckets by gender (18-24, 25-34, 35-44, 45-54, 55+)

Full detail additionally includes:

• Emotion distributions by gender per camera (ANGRY, DISGUST, FEAR, HAPPY, NEUTRAL, SAD, SURPRISE — totalConfidence / count = average confidence)
• Hourly detection/track/unique people patterns (24-element arrays, index = hour of day)
• Daily breakdowns with per-camera sub-detail

Use "summary" first to assess the facility, then "full" for deep analysis on specific areas.

Part of the AI Insights & People module.

Searches AI-powered face detection timeline blocks for a facility — the primary "who was in the facility" view.

Returns paginated, pre-aggregated visit blocks showing who was seen, when, for how long, and on which cameras. Backed by OpenSearch for fast queries.

Key capabilities:

• Filter by date range, membership status, match type (identified/unsure/blacklist)
• Search by person name using free-text query
• Filter by specific person ID
• Exclude or include specific camera zones (e.g., exclude "Reception" to avoid false positives)
• Paginated with facets for understanding data distribution without fetching all records

Detail levels (use the 'detail' parameter):

• "summary": personName, personId, matchType, membershipStatus, timestamps, duration, cameraCount (~200 bytes/result)
• "standard" (default): Adds cameras with names/groups, dominant emotion, age, dominantGender (~500 bytes/result)
• "full": Adds raw emotion confidence arrays per camera (~1.5-2.5KB/result). Emotion confidence: divide totalConfidence by count to get average.

Camera zone filtering:

• excludeCameraGroups: Exclude results where the person was ONLY seen on these camera groups. Useful for excluding Reception area (people with expired memberships often go to reception to cancel — not a genuine facility visit).
• includeCameraGroups: Only include results where the person was seen on at least one of these camera groups.

Facets returned:

• matchType, cameraIds, membershipStatus, isOngoing, checkInFound — with counts per value

Use this tool after insights_report_snapshots to investigate who specifically visited.

Part of the AI Insights & People module.