Skip to content
TecLeads
TelecommunicationsEMEAAI

Shipping a production RAG copilot for a telco's frontline support

Designed and operated a RAG assistant powered by an open-weights model on the customer's own infrastructure, with red-teaming and evaluation harnesses built in.

Client: An EMEA mobile network operator

Challenge

What we walked into

Frontline agents spent 40% of call time searching internal knowledge bases. Closed-model APIs were ruled out due to data sovereignty.

Approach

How we engaged

  • Stood up gpt-oss on Ollama / vLLM behind an internal API gateway with auth and rate limits
  • Built an evaluation harness with 800 golden cases and automated graders
  • Indexed 2M support documents into pgvector with hybrid search
  • Designed prompt-injection defenses, PII redaction, and audit logging
  • Integrated with the existing CCaaS platform via WebSocket
Results

What changed

Average handle time
−27%
First-call resolution
+18 pts
Cost per call (vs. closed-API alternative)
−72%

TecLeads understood that a model is only as useful as its evaluation harness. They built ours from day one.

Director of Customer Operations · Telecommunications

Looking for similar outcomes?

Tell us about your situation. We respond within one business day.

This case study is an illustrative, anonymized representation of recent client work.