Project 04 — RAG / AI agent

Ask your Drive.
Get a report.

n8nOpenAIGoogle Gemini embeddingsPineconeDrive · Sheets · DocsRole: design + build

Company knowledge lived in scattered Drive files — answering a question meant hunting documents by hand. I built a retrieval pipeline plus an AI agent: it ingests the corpus into Pinecone, answers from retrieved context, and writes the report straight to Google Docs.

PROBLEM

Dozens of source files, one shared Drive, zero recall. Every research question turned into 30+ minutes of open-tab archaeology — and reports were assembled by copy-paste.

PROCESS

Two branches, one system. Ingest: a Google Sheet lists files to load → loop over items → download from Drive → Recursive Character Text Splitter → Gemini embeddings → Pinecone Vector Store. Answer: an AI Agent (OpenAI chat model, with memory) gets a Vector Store Tool backed by Pinecone retrieval, and its output is saved via update-document to Google Docs. Retrieval-first means answers stay grounded in your files, not model imagination.

RESULT → Questions over the full corpus answered in seconds with a report auto-filed to Docs — research per report down from ~30 minutes to seconds.
Ground My Docs Like This →← All Work