Transform Excel data into AI-ready vectors with OpenAI and Supabase

Description This workflow is a practical, “dirty” solution for real-world scenarios where frontline workers keep using Excel in their daily processes. Instead of forcing change, we take their spreadsheets as-is, clean and normalize the data, generate embeddings, and store everything in Supabase.

The benefit: frontline staff continue with their familiar tools, while data analysts gain clean, structured, and vectorized data ready for analysis or RAG-style AI applications.

How it works - Frontline workers continue with Excel – no disruption to their daily routines. - Upload & trigger – The workflow runs when a new Excel sheet is ready. - Read Excel rows – Data is pulled from the specified workbook and worksheet. - Clean & normalize – HTML is stripped, Excel dates are fixed, and text fields are standardized. - Batch & switch – Rows are split and routed into Question/Answer processing paths. - Generate embeddings – Cleaned Questions and Answers are converted into vectors via OpenAI. - Merge enriched records – Original business data is combined with embeddings. - Write into Supabase – Data lands in a structured table (excelrecords) with vector and FTS indexes.

Why it’s “dirty but useful” - No disruption – frontline workers don’t need to change how they work. - Analyst-ready data – Supabase holds clean, queryable data for dashboards, reporting, or AI pipelines. - Bridge between old and new – Excel remains the input, but the backend becomes modern and scalable. - Incremental modernization – paves the way for future workflow upgrades without blocking current work.

Outcome Frontline workers keep their Excel-based workflows, while data can immediately be structured, searchable, and vectorized in Supabase — enabling AI-powered search, reporting, and retrieval-augmented generation.

Required setup

Supabase account - Create a project and enable the pgvector extension.

OpenAI API Key - Required for generating embeddings (text-embedding-3-small).