Zepto uses evaluation-first AI to scale support
Zepto, an Indian rapid delivery platform, now handles over 100,000 customer support tickets daily using a multi-agent AI system. Instead of just launching more agents, Zepto shifted to building, testing, and running these bots on Databricks with a strong focus on evaluation — constantly checking if they’re actually working. This approach aims to keep speed and reliability in balance as the company grows.
- Zepto handles over 100,000 support tickets daily
- AI agent errors can create thousands of bad outcomes
- Databricks powers the evaluation infrastructure
- Evaluation finds and fixes internal AI failures fast
Sources covering this
In this story
More in Enterprise
Red Hat adds NVIDIA BlueField to OpenShift
Red Hat OpenShift will now support NVIDIA BlueField hardware, aiming to make data centers faster by letting specialized cards take on tasks usually handled by main processors. By moving network traffic management and security enforcement off CPUs and onto BlueField, businesses can free up computing power for core applications and AI tasks. Red Hat says this helps cut operational costs and makes network security more robust.
Google Data Cloud adds new AI and analytics tools
Google just launched several new features for its Data Cloud. BigQuery now supports stateful processing in streaming queries, letting you do things like rolling averages in real time (currently in preview). There’s also a ready-to-use synthetic data generator for Managed Kafka, now generally available. And Lakehouse managed tables for Apache Iceberg are in preview, promising easier data management and cross-tool compatibility. Updates to Dataflow pipelines are also rolling out.
GitLab Duo Self-Hosted supports multiple AI models via Azure
GitLab is letting organizations use its AI coding features with their own choice of large language models, including OpenAI, Anthropic, Meta, and Mistral, when deployed through Microsoft’s Foundry platform on Azure. Companies can now pick which region, network, and cloud controls handle their code—with model assignments down to the individual feature. This helps teams meet data residency and compliance needs.
Salesforce launches new Agentforce bundles with built-in AI
Salesforce just unveiled three bundled versions—Core, Advanced, and Max—of its Agentforce tools for sales, service, and industry customers. These bundles now include built-in AI, Slack, upgraded analytics, enterprise-level security, and customer support, all for a single price. The Max edition, according to Salesforce, offers 60% more value than the previous Agentforce 1, with existing customers able to upgrade at no extra charge.