AI Dashboard Generation Platform
Built an AI system that transforms Excel data into branded HTML dashboards across 5 industry verticals. Through reverse-engineering existing outputs, we discovered a universal data schema and 4 processing patterns that power all dashboard components, enabling a scalable 6-stage pipeline architecture that can handle new verticals without custom development.
5
Verticals
17
Components
4
Patterns
6-stage
Pipeline
Background
Makam had successfully delivered custom data dashboards to clients across insurance, automotive, legal, financial services, and real estate. Each dashboard required significant manual effort to produce. They wanted to automate the process using AI but weren't sure how to architect a system that could handle the apparent diversity of their outputs.
Challenge
Makam had 7 working demo outputs across different industry verticals, but no scalable architecture. Each dashboard appeared unique in structure and content, making it unclear how to build a platform that could handle new verticals without custom development for each one. The challenge was finding the common patterns hidden beneath surface-level differences, then designing an AI pipeline that could reliably produce quality outputs at scale while keeping costs manageable.
Approach
Reverse-engineered all 7 demo outputs, documenting every component, data source, and transformation
Discovered that despite visual differences, all dashboards followed a universal 9-10 column Excel schema
Identified 4 fundamental processing patterns that power all components: Count, Classify, Select, and Synthesize
Mapped all 17 distinct dashboard components to their underlying processing patterns
Designed 6-stage pipeline: Excel ingestion, Brief interpretation, Per-row classification, Component assembly, QA validation, HTML rendering
Implemented smart model allocation routing bulk classification to fast/cheap models while reserving expensive models for synthesis
Built component configuration system allowing new verticals to be added through configuration rather than code
Deliverables
Universal data schema documentation explaining the 9-10 column structure all verticals share
Processing pattern specifications with examples showing how each pattern transforms data
Complete 6-stage pipeline architecture: Excel to Brief to Classification to Components to QA to HTML
Working prototype demonstrating the full pipeline on sample data from multiple verticals
Component configuration system documentation for adding new dashboard types
Cost optimization analysis showing expected per-dashboard costs with tiered model allocation
Results
Insurance, automotive, legal, financial, and real estate all supported
Distinct dashboard components identified and mapped to processing patterns
Universal processing patterns: Count, Classify, Select, Synthesize
Production-grade architecture from raw data to finished dashboard
Impact
The architecture reduced per-dashboard production time from days to hours. More importantly, adding new verticals now requires configuration changes rather than custom development, fundamentally changing Makam's unit economics and scalability.