OptimaAI Logo

Pages & Features

Dashboard, Predictions, Recommendations, BMC and Knowledge Base pages

Pages & Features

Dashboard (Home)

The main dashboard provides a real-time overview of the business's health.

Components:

  • KPI cards — Revenue MAPE, Churn ROC-AUC, Growth R², Prophet MAPE
  • Revenue forecast chart — 12-period projected revenue from the ML model
  • Growth trend chart — 12-month growth projection
  • Churn rate gauge — current churn % with risk colouring

Data source: GET /api/v1/kpis


Predictions

The Predictions page lets users run individual forecasts on demand.

Revenue Prediction form

Inputs: quantity_sold, price, discount_percent, seasonal_index, gross_margin_pct, month, quarter

Output: predicted net revenue in USD

Churn Prediction form

Inputs: days_since_last_order, total_orders, avg_days_between_orders, churn_risk_score, total_spent, avg_order_value

Output: churn probability + colour-coded risk badge (low / medium / high)

Growth Prediction form

Inputs: same feature shape as revenue

Output: 3-month forward revenue forecast


Recommendations

The Recommendations page generates an LLM-written briefing for the selected role.

  • Role selector: Sales Manager / Finance Controller / Executive
  • Submitting fetches from POST /api/v1/recommend
  • Response is rendered as formatted markdown
  • Results are cached — the same role won't trigger a new LLM call for 1 hour

Business Model Canvas

The BMC page generates and displays the full 9-block canvas.

Blocks rendered:

  1. Customer Segments
  2. Value Propositions
  3. Channels
  4. Customer Relationships
  5. Revenue Streams
  6. Key Resources
  7. Key Activities
  8. Key Partners
  9. Cost Structure

Each block is rendered as a card with bullet points parsed from the LLM response. The BMC is cached for 24 hours; clicking "Regenerate" clears the cache and triggers a new LLM call.


Datasets

The Datasets page manages uploaded business data files.

Features:

  • File upload button (accepts CSV, Excel, PDF, TXT, JSON)
  • Upload triggers the full preprocessing pipeline in the backend
  • Table listing all uploaded datasets with filename, rows, quality score, and category
  • Delete button removes the record and cleaned table

Knowledge Base

The Knowledge Base page provides a chat-style interface for querying ingested documents.

Features:

  • Free-text query input
  • Fetches top-3 relevant document chunks from ChromaDB via RAG
  • LLM generates a grounded answer citing the retrieved context
  • Source chunks displayed below the answer for transparency

On this page