How Kimola Leverages Llama to Secure Text Analysis Data for Enterprises

Kimola is a customer feedback analysis platform trusted by companies in over 90 countries from Michelin Brasil to Costa Coffee UK, Honda Netherlands to P&G Singapore to analyze customer feedback, customer tickets, survey responses, chats, and call center conversations across 30+ languages. At the heart of its solution lies proprietary Natural Language Processing (NLP) technology that performs auto-classification and thematic analysis—no manual training required.

For deeper insight, Kimola integrates with large language models (LLMs) to interpret this analyzed data. That includes generating executive summaries, extracting feature requests, or surfacing customer pain points based on the structured insights produced by Kimola’s engine, using LLM's.

But what happens when an enterprise client requires all of this—without any data leaving their internal systems?

The Data Privacy Challenge

In 2024, Deswik, a subsidiary of a NASDAQ OMX-listed company, approached Kimola with a critical requirement: an on-premises solution. Due to strict data privacy and confidentiality requirements, Deswik could not send customer support data to any cloud infrastructure—including LLM APIs. Moreover, their project involved a large-scale assessment of their historical tickets, making usage-based pricing models impractical.

A Hybrid, On-Prem Solution

At the time, Kimola had just been accepted to the Meta LLM Startup Program, a global initiative supporting startups working with large language models. This opened the door for Kimola to experiment with integrating open-source LLMs into privacy-focused deployments.

For Deswik, this meant deploying an offline LLM alongside Kimola’s proprietary analysis engine.

Here’s how the solution was structured:

• Kimola’s engine analyzed the raw text data—auto-classifying topics, clustering issues, and identifying themes.
• The offline LLM interpreted those results—generating natural language summaries, pain point narratives, and internal reports.

“Data security is a core principle for us at Deswik,” says Project Lead, Yu Nakagawa. 

“Kimola offered us an offline LLM deployment as part of their on-prem solution, allowing us to keep our customer tickets fully secured. Over just three months, we were able to analyze tens of thousands of tickets with zero compromise on security and privacy.”

Why This Works

By separating analysis and interpretation, Kimola created a powerful hybrid setup that worked within Deswik’s strict constraints:

• Scalable and private: Kimola’s analysis engine ran efficiently and securely on-prem, processing massive volumes of data with no cloud dependency.
• Interpreted insights without the cloud: The offline LLM provided all the language understanding and summarization value—without transmitting a single word externally.
• Fixed costs: No token-based pricing. Deswik could scale usage without worrying about unpredictable bills.

The Open Advantage

By opting for an open-source LLM, Kimola enabled Deswik to benefit from:

• Full transparency into how the LLM was being used and interpreted.
• Full security, with no reliance on external infrastructure.

Setting a New Standard for Enterprise AI

Enterprise AI demands the balance of deep insights and robust security—and Kimola delivers just that. By leveraging its custom analysis engine alongside an offline LLM interpreter, Deswik was able to analyze a large volume of tickets within just three months—without transmitting a single one to the cloud.