hanen-gb/etl-n8n-agent

ETL process automation agent (n8n) - Sofrecom Tunisia internship # ETL Process Automation Agent — Sofrecom Tunisia Internship Agentic conversational agent (n8n) that automates end-to-end Excel file processing: selecting a source file from Google Drive, applying real transformations (cleaning, formatting, dates, deduplication, calculated columns, pivot tables) on a **working copy** Google Sheet, answering questions/statistics computed deterministically (no numbers invented by the LLM), and final multi-destination loading (Local, Google Drive, SharePoint). Project built as part of an internship at **Sofrecom Tunisia**. ## Table of contents - Overview - Architecture - Repository contents - Quick start - Usage - Known limitations - Detailed documentation ## Overview The user chats with the agent (n8n chat or webhook) in natural language. A router agent classifies each message into an intent, then routes to the matching sub-flow: | Intent | What it does | |---|---| | `select_file` | Lists the raw Excel files available in Drive, or — if a name is given — creates a working copy and converts it to a Google Sheet | | `transform` | Applies a real transformation on the working copy: text cleaning (upper/lower/proper case), date normalization, deduplication, calculated column, cell formatting, or pivot table | | `question` | Answers a numeric question (count, sum, average, distinct values...) using statistics computed in JS on the real data — the LLM only phrases the answer | | `visualisation` | Proposes a chart configuration from the same real statistics | | `consult` | Returns the working-copy Google Sheet link at any time, for visual verification | | `loading` | Exports the working copy to a clean `.xlsx` and delivers it to 3 destinations: local disk, `Final` Drive folder, SharePoint | | `sharing` | Prepares a share message (email, Slack, Teams...) | | `conversation` | Regular conversational exchange | Each session's state (selected file, current working copy, status) is kept in a Postgres table (`etl_sessions`), so a conversation …