OgaMetrics AI
A multi-tenant analytics and RAG platform that turns raw business data into searchable AI insights. Automates ingestion, cleaning, embeddings, and vector search to deliver fast, intelligent answers for SMEs and enterprise.
THE PROBLEM
SMEs and enterprise teams sit on large amounts of structured business data (CSV exports, spreadsheets, reports) but lack the infrastructure to query that data intelligently. Analytics platforms are expensive and require data engineering expertise.
THE PRODUCT
OgaMetrics AI is a multi-tenant platform that automates the full data pipeline from raw CSV ingestion through cleaning, schema detection, normalization, embedding generation, and vector indexing. Once data is indexed, users can ask natural-language questions and receive AI-generated answers grounded in their specific business data.
ENGINEERING DECISIONS
- 01
FastAPI backend chosen for its async-first design and native Python AI/ML library compatibility
- 02
Multi-tenant architecture with tenant-scoped vector stores for data isolation
- 03
Schema detection runs heuristics to auto-identify data types, reducing manual configuration
- 04
RAG architecture grounds LLM responses in tenant-specific data, preventing hallucination
- 05
Embeddings generated per-chunk for fine-grained retrieval quality
CHALLENGES
- ›
Handling messy real-world CSVs: mixed types, missing values, inconsistent headers, encoding issues
- ›
Keeping vector search fast as data volumes grow without expensive infrastructure
- ›
Ensuring tenant data isolation in a shared vector index
- ›
Evaluating retrieval quality — knowing when RAG is returning relevant vs. irrelevant context
RESULTS
- ✓
Live demo deployed at ogametrics.vercel.app
- ✓
Full pipeline from CSV upload to AI query working end-to-end
- ✓
Multi-tenant data isolation implemented with scoped vector collections
WHAT I WOULD IMPROVE
- →
Add retrieval evaluation metrics to measure RAG quality per tenant
- →
Support Excel and JSON in addition to CSV inputs
- →
Build an agentic layer for multi-step analytical workflows
- →
Add streaming responses for faster perceived performance on large datasets