AI agents that absorb your repetitive customer support.
On WhatsApp and connected to your CRM and ERP, they triage, handle billing and work orders, and hand off to a human with the full history. For internet providers and service companies with heavy support volume.
- Led 70+ people and 12 engineers
- Salesforce: 3 years in production
- Brasília, Brazil · remote
FIG.01Pedro V. ZimmermannWhat I built and what changed
At Explorernet, a telecom with nearly 100k subscribers, where I was CTO and managing partner from 2019 to 2025.
Customer support run by AI agents
Problem
Expensive, reactive support that depended on 70 people.
What I built
I co-created TalkToMe and put AI agents on the official WhatsApp channel, connected to the CRM. They triage, resolve and escalate to a human with the full history. Billing, work orders and omnichannel were wired to the agents.
Measured result
Salesforce as the backbone of operations
Problem
An ERP that needed manual work and parallel spreadsheets.
What I built
I rolled out Salesforce from scratch, replaced spreadsheets with a direct ERP integration and connected billing, work orders and WhatsApp to the AI pipeline.
Measured result
Network diagnostics by geolocation
Problem
Failures were only found after customers complained.
What I built
Real-time monitoring that cross-references network events with their location.
Measured result
Agents in production,
not prototypes.
Short scope, in writing
We agree on what the agent will solve and how you will approve it. The first version runs on mock data, and dependencies (Meta approval, ERP access) are explicit from day one.
Tests on the real channel before go-live
An end-to-end test suite checks handoffs between agents, tool order and leaks of internal content. Latest recorded run: 6 of 6 critical scenarios at two providers (Jun 2026).
Cost of every conversation, measured
Measured example: R$ 0.018 (Brazilian reais) for a conversation with an agent handoff (Jun 2026). I compare models side by side: in an Apr 2026 case, the smaller model was 4× cheaper per session with no loss in quality (15 scenarios × 4 models).
Human in the loop
Queue, supervision and history in one panel. The agent hands off to your team when it needs to.
You only pay after you approve: I build the agreed scope, you test it working, then you decide.
From network cables to CTO
I joined Explorernet in 2009 as a maintenance technician, swapping hardware and pulling network cable. I came back in 2019 as CTO and managing partner and stayed until 2025, leading more than 70 people, including a team of 12 engineers.
Today I deploy customer-support agents on TalkToMe for other companies, including internet providers.
Stack: Python · TypeScript · OpenAI · Anthropic/Claude · Gemini · Supabase (pgvector) · Docker · Kubernetes · Salesforce
Let's look at your customer support
Send me one WhatsApp message with: what your company does, what you want AI to solve, how many conversations you get per month and which systems it would need to access.
Hiring an AI Tech Lead or Tech Lead? Reach me by email or LinkedIn. I work 100% remote or in Brasília, Brazil.