Cluma engineering AI project.

May 2026

01/The project

For my industry project, I developed Cluma Data Matching, a full-stack proof of concept (PoC) asset data enrichment tool designed to automate the extraction, verification, and completion of technical asset information stored in Odoo ERP. I built a web application that automatically crawls network documents, extracts relevant technical data using AI-assisted parsing, and compares it with existing Odoo records via a secure human-in-the-loop validation workflow to prevent accidental data overwrites.

Tech stack

Vue

Tailwind

Docker

Python

Figma

Roles

FullstackDesign

Cluma engineering AI project project hero

02.The challenge

In many companies, technical asset data like serial numbers and electrical specifications is fragmented across isolated network shares, PDFs, emails, and spreadsheets. This fragmentation leads to incomplete, disjointed ERP records and manual data-entry bottlenecks.

03.The implementation

The backend uses Python and FastAPI to ingest data from PDFs, Word, Excel, .msg files, and images, using Tesseract OCR as a fallback. To optimize performance, text chunks and vector embeddings are stored via PostgreSQL and pgvector. The system integrates gpt-4o-mini for structured JSON proposals, using strict backend guardrails to cross-verify outputs against raw source text to eliminate hallucinations.

The frontend is a responsive Vue 3, TypeScript, Pinia, and Tailwind CSS v4 dashboard featuring asset views, live file previews, and an approval interface. The entire ecosystem is containerized using Docker and Docker Compose.

04.The result

This project successfully bridged traditional software engineering with modern AI, providing deep experience in handling inconsistent real-world data and implementing deterministic guardrails around LLM outputs.