Production RAG platform

Multi-tenant RAG platform for BCRP

A multi-tenant RAG platform that brings grounded search and assistant experiences to four BCRP applications.

Role

Full Stack AI Engineer / GenAI & RAG Developer at Canvia

Stack

Python · FastAPI · LangChain · Node.js · Oracle 23ai · pgvector · Pinecone · Chroma · OpenAI

Public BCRP Research Portal homepage showing the research navigation and assistant button. — Public capture of the BCRP Research Portal. The case study keeps implementation details high-level and links to the live product.

Problem

Several institutional applications needed reliable search and assistant experiences over thousands of research documents.

My contribution

At Canvia, I designed, built and improved the platform, including ingestion and indexing, shared tools, configurable chat/search/prompt modules, multimodal ingestion, and daily testing with human review.

Architecture & stack

Node.js gateway, Python backend, Oracle 23ai, pgvector, Pinecone, Chroma, SQL pipelines to RePEc, and multimodal document processing.

Result

One platform serving four BCRP applications; its assistant is live on the Research Portal.

Evidence-led delivery

I worked across ingestion and indexing, shared tools, configurable chat, search and prompt modules, multimodal document processing, and daily testing with human review. The public story focuses on the engineering contribution without exposing institutional documents or internal data.

A public-safe boundary

The architecture sketch is intentionally high-level: it shows the flow from institutional sources through retrieval and shared services to the applications, while leaving out credentials, private endpoints and operational details.

From documents to grounded answers

Sources & ingestion

Research documents, multimodal parsing and reproducible indexing.

Retrieval & services

Node.js gateway, Python services and configurable RAG modules.

Applications

Four BCRP applications, including the live Research Portal assistant.