Briefcase — Client GPT

Creator & Architect AI Systems RAG Pipeline MCP
Briefcase system architecture
System architecture diagram — MCP pipeline connecting Asana, Google Drive, and multiple LLM interfaces

Overview

An MCP-based RAG pipeline that unifies the agency's Asana projects, Google Drive, meeting notes, media plans, and websites into any LLM interface — enforcing consistent, high-quality output across 215+ client accounts and 60+ individual contributors. Built solo, from PRD through production rollout.

The Problem

Knowledge was fragmented. An agency-wide survey I ran revealed that 61% of participants identified "centralize client knowledge" as their top pain point. Context got lost between client handoffs. Past decisions lived in emails, meeting notes, and individual drives. Teams were making avoidable mistakes — not because they lacked skills, but because the information wasn't accessible.

Approach

I chose MCP (Model Context Protocol) as the architecture so the system could work inside any LLM interface the team already uses — Claude, ChatGPT, Gemini, whatever comes next. The tool meets people where they are instead of introducing yet another product to manage.

Supabase stores vectorized project data. Railway hosts the pipeline. Individual access is granted per user. I ran enablement sessions with ICs on connecting Briefcase to their existing third-party stacks — Webflow, SEMRush, ad platforms.

Process

  1. Surveyed 60+ team members across every department to map knowledge pain points
  2. Synthesized findings — fragmented context was the root cause of inconsistent output
  3. Chose MCP architecture so the tool works regardless of which LLM each person prefers
  4. Solo-built the full pipeline — PRD writing, architecture planning, and iterative build cycles
  5. Loaded 215+ client accounts from Asana and Google Drive automatically on first run
  6. Implemented beta rollout with feedback tooling baked in for rapid iteration
  7. Running ongoing enablement sessions with ICs on usage and third-party connectivity

Design Artifacts

Process artifacts will be added as this case study expands.

Results

215+

Client accounts loaded

60+

Survey participants

1

Solo builder

LLM-agnostic by design via MCP. Production-ready, deployed agency-wide. Enables consistent output without changing anyone's existing workflow.

Let's talk.

Currently at The Influence Agency, Toronto. Open to conversations about product design, design systems, and design leadership.

roccobmb@gmail.com / LinkedIn / Resume / GitHub