Formlabs Sales & Services AI Agents
Dual AI agents (Rhys for Sales, Pete for Services) built with Streamlit and Gemini. Supports RAG via Vertex AI, Google Search fallback, Zapier lead/case creation, and BigQuery-powered printer log lookups.
Context
Built during a Formlabs-sponsored hackathon to streamline inbound sales qualification and service case intake. The solution replaces fragmented, manual workflows with a guided conversational flow powered by Gemini and company knowledge.
The Problem
Manual triage and data capture across sales and support led to slow response times and inconsistent handoffs. Agents lacked a unified interface that could both answer questions and structure customer data for downstream systems.
Approach
Implemented two focused chat agents in Streamlit: Sales (Rhys) and Services (Pete). Used Gemini with Vertex AI RAG or Google Search, instructed the model to emit structured JSON for lead/case data, then posted to Zapier webhooks. For services, integrated a BigQuery lookup to surface recent print jobs by printer serial.
Architecture
Python/Streamlit UI (`app.py`) orchestrating backends in `sales_backend/` and `services_backend/`. Sales: Gemini (Vertex AI RAG or Google Search) with JSON extraction and Zapier lead creation. Services: Gemini with JSON extraction, Zapier case creation, and BigQuery client for recent job queries. Environment via dotenv; dependencies in `requirements.txt`.
Results
Automatically captures qualified leads from chat to Zapier with budget, timeline, and email
Creates services cases and links follow-up via Zapier; looks up recent jobs in BigQuery by printer serial
Switchable modes: Vertex AI RAG vs. Google Search for sales; services always on Search
Streamlit chat UX with avatars, typing state, and session history for rapid internal rollout
What I Learned
Refined prompt/system-instruction design for reliable JSON emission; balanced RAG vs. web search; integrated BigQuery in conversational flows; handled Streamlit reactivity/state to keep chat responsive and robust.
Impact Summary
Automates sales qualification and service triage; captures qualified leads to Zapier and opens service cases with BigQuery context. Built at a Formlabs-sponsored hackathon and now being implemented by Sales and Services.