AI Dashboard Generation Platform
Built an AI system that turns raw Excel data into branded HTML dashboards across 5 industry verticals. By reverse-engineering existing outputs we discovered a single universal data schema and four core processing patterns beneath the surface variation, then designed a scalable six-stage pipeline with cost-tiered model routing, so new verticals are added by configuration, not custom code.
Hours
Production
5
Verticals
4
Patterns
6
Stages
Background
Makam delivered custom data dashboards to clients across insurance, automotive, legal, financial services, and real estate. Each dashboard required heavy manual effort to produce. They wanted to automate the process using AI but weren't sure how to architect a system that could absorb the apparent diversity of their outputs at scale.
Challenge
Makam had a portfolio of demo outputs across different industry verticals, but no scalable architecture. Each dashboard looked unique in structure and content, making it unclear how to build a platform that could handle new verticals without bespoke development. The challenge was finding the universal patterns hidden beneath surface variation, then designing a pipeline that could reliably produce branded, quality-checked outputs at scale while keeping per-dashboard costs predictable.
Approach
Reverse-engineered the existing demo outputs, documenting every component, data source, and transformation
Discovered a single universal Excel schema that all verticals shared despite visual differences
Identified four core processing patterns powering every component: Count, Classify, Select, and Synthesize
Mapped 17 distinct dashboard components to their underlying processing patterns
Designed a six-stage pipeline: ingestion, brief interpretation, per-row classification, component assembly, QA validation, and HTML rendering
Implemented cost-tiered model routing so bulk classification runs on fast, low-cost models while expensive models are reserved for synthesis
Built a component-configuration system so adding a new vertical becomes a config change rather than a development project
Deliverables
Universal data schema documentation describing the column structure every vertical shares
Processing-pattern specifications with examples showing how each pattern transforms data
Six-stage pipeline architecture from Excel ingestion through QA validation to branded HTML rendering
Cost-tiered model routing strategy with per-stage model allocation and budget projections
QA validation stage that checks each output against the brief before publication
Vertical-as-configuration system enabling new industries without code changes
Results
Per-dashboard production dropped from days of manual effort to hours of automated pipeline runs
Insurance, automotive, legal, financial services, and real estate all supported by the same pipeline
Universal processing patterns (Count, Classify, Select, Synthesize) power all 17 components
Ingestion, brief interpretation, per-row classification, component assembly, QA validation, rendering
Impact
Per-dashboard production dropped from days to hours, and adding a new industry vertical became a configuration change rather than a development project. That is a fundamental shift in unit economics, letting Makam scale its delivery without proportionally scaling its team.