How googlecloudplatform/terraform-genai-knowledge-base works
This Terraform-deployed knowledge-base solution turns uploaded documents into validated fine-tuning data and a deployed model. The main runtime path is document upload → event-driven webhook → OCR, indexing, and question/answer generation → Firestore-backed dataset export. README-described fine-tuning, human validation, model deployment, and prediction capabilities are included; their internal implementation details are not established by the sampled source.
Main entry points: main.py Event webhook · mim Model fine-tuning · main.tf Cloud resources · model-predictions.ipynb Model predictions
Components 15
Cloud deployment
- Cloud resources
main.tf
Document processing
- Event webhook
webhook/main.py - Document OCR
webhook/main.py - Page text
webhook/main.py - Page embeddings
webhook/main.py - Vector Search
- QA generation
webhook/main.py - Firestore records
Training and inference
- Dataset export
webhook/main.py - Tuning dataset
- Human validation
- Model fine-tuning
mim - Model Registry
- Model predictions
notebooks/model-predictions.ipynb
Other components
- Document uploader
Connections 15
- Cloud resources → Event webhook: deploys
- Document uploader → Event webhook: triggers
README.md - Event webhook → Document OCR: requests OCR
webhook/main.py - Document OCR → Page text: returns text
webhook/main.py - Event webhook → Page embeddings: requests embeddings
webhook/main.py - Event webhook → Vector Search: indexes pages
webhook/main.py - Event webhook → QA generation: generates Q&A
webhook/main.py - Event webhook → Firestore records: stores records
webhook/main.py - Event webhook → Dataset export: exports dataset
webhook/main.py - Dataset export → Firestore records: reads records
webhook/main.py - Dataset export → Tuning dataset: writes JSONL
webhook/main.py - Tuning dataset → Model fine-tuning: supplies data
README.md - Human validation → Model fine-tuning: validates data
README.md - Model fine-tuning → Model Registry: deploys model
README.md - Model Registry → Model predictions: serves predictions