Powering safer, smarter operations with AI and IoT

telsen, telsen, telsen, telsen, telsen, telsen, telsen, telsen, telsen, telsen, telsen, telsen, telsen,

Product strategy
AI design

Claude HaikuAI development

Introduction

Telsen is an IoT monitoring platform helping organisations across hospitality, food production, retail and facilities management monitor compliance, reduce operational risk and improve efficiency.

Customers previously had to navigate multiple screens and reports to answer everyday questions such as “are any fridges out of range?” or “did we miss any checks yesterday?”. Studio Graphene worked with Telsen to embed a conversational assistant directly into the platform, giving users a plain-English way to interrogate their own operational data.

The assistant inherits Telsen’s existing permissions and company-level access controls, so every answer is automatically scoped to the user, their role and the sites they are allowed to see.

For common questions, we created five quick actions covering areas such as open alerts, today’s checks, overdue workflows and connectivity issues. These bypass the language model entirely and call Telsen’s underlying tools directly, returning deterministic answers with no AI cost or external data egress.

Solution

Studio Graphene built and deployed a read-only AI assistant in production across Telsen’s web and mobile products, using the platform’s existing MCP tool layer and Claude Haiku LLM model to access data on sensors, devices, alerts, HACCP checks, workflows and sites.

The assistant inherits Telsen’s existing permissions and company-level access controls, so every answer is automatically scoped to the user, their role and the sites they are allowed to see.

For common questions, we created five quick actions covering areas such as open alerts, today’s checks, overdue workflows and connectivity issues. These bypass the language model entirely and call Telsen’s underlying tools directly, returning deterministic answers with no AI cost or external data egress.

For free-text questions, the assistant uses an agentic loop across a set of approved read-only MCP tools, translating natural-language requests into the right data queries and returning the answer conversationally. Users can ask follow-up questions within the same session without having to repeat the context, turning complex operational journeys into simple exchanges.

The architecture also supports per-company model routing. Claude Haiku 4.5 is used by default for cost efficiency, while GPT 4o mini can be enabled for individual companies, chosen in practice for its more consistent formatting and instruction-following compared with the default model. Sessions remain pinned to the same provider throughout a conversation, giving Telsen flexibility without changing the experience for the end user.

Results

83% of logged-in customers interacted with the AI assistant on Day 1 and made NLP queries on their dashboard data.
55% of customers made cross-statistics analytical queries, which was not possible earlier.
32% of customers logged positive feedback on Telsen, praising the new assistant and how it is helping them get insights beyond a dashboard.

The result is a conversational layer that turns Telsen’s operational and compliance data from something users have to search for into something they can simply ask about. Teams can surface alerts, workflow status, missed checks and site conditions in plain English, while the underlying role-based permissions and read-only controls remain intact.

pulsenext project