The all-new Chinese edition of the MWC Series app now available.

Close

GSMA Programmes

Open Telco AI: Build Agentic Evaluations for Telecom - Workshop & Hackathon

Open to All

Date

Thu, 25 Jun

Time

15:00 - 18:00 CST

Doors Open

14:50 CST

Location

Level 1, The BREW, Kerry Hotel

Presented By

Enrique Molero, AI Research Lead, Zhang Le China Telecom, GSMA, Louis Powell, Director of AI initiatives GSMA Technology and Hong Liu, Head of Technology, Greater China, GSMA.


Loading…

Session Description

Telecom needs its own SWE-Bench. Spend an afternoon building it. 

A three-hour hands-on workshop and hackathon co-hosted by GSMA and China Telecom at MWC Shanghai 2026. Teams turn real China Telecom use cases into agentic evaluations; a panel of judges picks the winners. 

Generic AI benchmarks are multiple-choice questions; picking the right answer based on telecom knowledge is a long way from operating a real network. Agentic evaluations are tasks where an agent does the actual work and gets graded on the result. This session is where telecom’s benchmark gets built. 

What you’ll do 

You work in a team of four to five. Your team logs into the Open Telco submission app and picks one of the real use cases contributed by China Telecom: multi-domain fault localization, say, or automated security alert triage. You then recreate that use case as a task family: a set of graded tasks that test whether an AI agent can actually do the work. You get two tools: 

  • Rowter, an experimental agent that gives you access to the telecom knowledge created in Open Telco; ask it what you would ask a domain expert. 
  • The Art of Evals, the guide on how to build agentic evaluations, written by the session’s main speaker, Enrique Molero. 

Mentors from GSMA, China Telecom and Open Telco AI circulate throughout. Submissions close at 18:00; the judging panel scores every task family and the top three teams present. 

Prizes 

A £3,000 cash pool across the top three teams, plus five MWC Shanghai 2027 passes for the winning team. 

Who should come 

The best evals pair operator depth with AI skill, so we want teams that mix both: network engineers, OSS and BSS specialists, telecom developers, AI and ML engineers, researchers, students and operator domain experts. 

Bring a laptop with a browser; there is nothing to install. If you register as an individual we will match you with a team on the day. 

What you leave with 

You leave knowing how to design agentic evaluations for telecom. 

Also in the room 

The day closes with the awards ceremony for the Telco Troubleshooting Agentic Challenge, a virtual competition hosted on Zindi. 

Agenda 

Time 

Block 

Detail 

Duration 

15:00 – 15:45 

Open Telco AI: what we are building 

The Art of Evals, walked through live: the structure of an evaluation; use case, agent, environment, suite, task; ending on a worked factory private 5G example 

45 min 

15:45 – 16:00 

Team formation & challenge brief 

Form teams of four to five; mentors join the tables; the submission app opens. Each team picks a China Telecom use case and writes a task family: agent, environment, suite 

15 min 

16:00 – 18:00 

Workshop + hackathon 

Based on the use case, teams build their task family end to end. Peer-feedback checkpoint at 17:00; submissions close at 18:00 

120 min 

18:00 – 18:30 

Telco Troubleshooting Agentic Challenge prize giving 

While the judging panel scores the submissions, the awards ceremony for the virtual challenge begins 

30 min 

18:30 – 19:00 

Hackathon winners & prize giving 

Top three teams present and prizes are awarded 

30 min 

Doors open: 14:50 



Doors Open: 14:50 CST

Session Speakers


Similar Sessions