Sanwa Comtec: AI-Driven Harassment Detection with IBM

Sanwa Comtec, an innovative IT solutions provider, specializes in crafting bespoke software packages that empower businesses to conquer new markets. Their flagship offering, the OSS "Rocket.Chat," delivers secure, on-premise chat communication solutions, catering specifically to clients with stringent cloud environment restrictions. Recognizing the growing need for enhanced security and ethical communication, Sanwa Comtec embarked on a mission to develop an AI-powered solution capable of detecting and flagging potentially harassing conversations within these secure, isolated environments.

To achieve this ambitious goal, Sanwa Comtec partnered with IBM to leverage the power of IBM® watsonx.ai™. Collaborating with legal experts, they meticulously curated training data comprising diverse examples of sexual and power-based harassment. This comprehensive dataset was then used to train a custom language model within watsonx.ai, harnessing the advanced classification capabilities of the IBM Watson® natural language processing (NLP) function. The resulting custom model is adept at identifying and flagging potentially inappropriate conversations in real-time.

The collaboration between Sanwa Comtec and IBM successfully demonstrated the potential of deploying lightweight language models within on-prem packages without relying on cloud-based LLMs. This is particularly crucial for organizations with strict data privacy and security requirements that prevent them from utilizing cloud-based AI solutions. By leveraging watsonx.ai, Sanwa Comtec was able to create a powerful, on-premise harassment detection system that met the unique needs of its clients.

Sanwa Comtec continues to collaborate with IBM to refine the application and enhance the accuracy of the AI model. This ongoing partnership ensures that the solution remains at the forefront of harassment detection technology, providing businesses with a robust tool to foster a safe and respectful communication environment.

Value Results:

• Enhanced Security: Securely deploy harassment detection within closed networks, addressing cloud restrictions.

• Improved Accuracy: Achieve over 95% accuracy in AI-based harassment judgments through continuous model refinement.

• Cost-Effective Solution: Leverage lightweight language models on-premise, reducing reliance on costly cloud-based GPUs.

• Scalable Implementation: Facilitate the deployment of AI-powered solutions in environments where uploading data to the cloud is restricted.