User Interface

Technology Stack

Signal™ — Technical Specification

Query Flow

CEO types query in Signal™ input
  → POST /api/query { query, customer_id, session_id, scope_override? }
  → Vectorized retrieval: pgvector similarity search on Aurora (RLS-scoped to customer)
  → Context window assembled from top-k retrieved records
  → Inference call (self-hosted vLLM) with system prompt + context + query
  → Streaming response via SSE: token-by-token to Signal™ UI
  → Each exchange turn written as JSON to Aurora session record
  → Markdown rendered in UI as stream completes

Session Persistence

Session records are written to Aurora on a per-turn basis — not at session end. This prevents data loss on browser close or network interruption. Each turn record contains the CEO query, the full model response, the retrieved context sources, and a timestamp.

signal_sessions table (Aurora, RLS-scoped)
────────────────────────────────────────────
session_id       UUID
customer_id      UUID (RLS tenant)
user_id          UUID (CEO user)
created_at       TIMESTAMPTZ
last_active_at   TIMESTAMPTZ

signal_turns table (Aurora, RLS-scoped)
────────────────────────────────────────────
turn_id          UUID
session_id       UUID (FK)
customer_id      UUID (RLS tenant)
turn_index       INTEGER
query            TEXT
response         TEXT
context_sources  JSONB (array of source references)
model_used       VARCHAR
created_at       TIMESTAMPTZ

Default Behavior & Overrides

Archive — Technical Specification

Archive is a read-only view of all Signal™ sessions. It is built from the signal_sessions and signal_turns tables in Aurora, filtered to the authenticated CEO's user ID and customer RLS tenant.

• Reverse-chronological session list — sorted by last_active_at DESC
• Session preview — first query of each session, truncated at 80 characters
• Infinite scroll — paginated query with cursor-based pagination
• Session detail view — full read-only conversation thread, markdown rendered
• Model read access — the inference layer can query Archive for cross-session continuity via the same RAG retrieval path, scoped to the CEO's session history

Admin Panel — Technical Specification

Employee Roster

The roster is a JOIN across the employee_consent table and a connector-status view, grouped by employee and pivoted by channel. Each cell reflects the current enrollment and sync state for that employee × channel combination.

Connector Health Dashboard

Connector health is derived from the connector framework's health monitoring layer. Each connector reports its last successful poll timestamp and current circuit-breaker state to a shared status store (DynamoDB). The Admin panel reads this store and renders status in real time.

connector_health table (DynamoDB, keyed by customer_id + platform)
──────────────────────────────────────────────────────────────────
customer_id          STRING (partition key)
platform             STRING (sort key)
status               STRING (green | yellow | red)
last_successful_poll ISO8601 timestamp
last_error           STRING (nullable)
error_count          NUMBER
stale_alert          BOOLEAN (true if last_successful_poll > 12 hrs ago)

Admin Actions

• Re-send enrollment link — generates a new UUID token for the employee (previous token expires); sends via admin's distribution method
• Add employee — creates a new employee record and enrollment link; employee added to all applicable connector sync scopes
• Remove employee — marks employee as inactive; consent record preserved for audit; communications data retention per customer data policy

Role-Based Access Control

Role assignment is stored in the users table in Aurora (RLS-scoped to the customer tenant). Route-level guards in the React application enforce role restrictions on the client. API endpoints enforce the same restrictions server-side — client-side guards are UI convenience only.

Full Feature Set