01The challenge

A year ago a telecom outage in Canada took down payment processing at banks and retail stores, plus consumer internet and mobile data. A national Tier 1 operator started looking for AI-based troubleshooting.

Their senior network operations team had more error logs arriving daily than it could read.

    Two lines in twenty-two carry the fault.

    02What we proposed

    An open-source data pipeline feeding ML and GenAI, streamed into an AI assistant an engineer can question directly.

    We built the platform around human-centric design and modeled a day in the life of these senior engineers.
    The design principle everything else was measured against

    We tested several leading LLMs and engineered the prompts that returned the best answers.

    03Widening the AI

    Error logs turned out to be a rich source rather than exhaust. So we layered ML on top of the raw stream.

    • Syslog formats from leading routers and switches
    • Firewalls and load balancers
    • Anomaly detection across the incoming stream
    • Pattern-based correlation between systems

    04Pilot to production

    After a successful pilot, the platform is going into their production network. We kept building, adding advanced dashboards and early warning insights.

    Monitoring tells you something broke. It does not tell you why. That gap is what the dashboards fill, alongside whatever alerting a NOC already runs.

    05The application stack

    We are adding the whole application stack to our log format library, servers and APIs included.

    We continue to focus and innovate so that our platform can scale to thousands of devices and millions of logs per day.
    The standing constraint on every release

    06Secure deployment

    The platform runs in secure on-premise data centers and in your private cloud. Sovereign-ready and cloud-agnostic, against only your own data sources.

    Back to all articles Read the telco case study