Case Study: Vannevar Labs achieves faster, more accurate multilingual defense intelligence with Databricks

Bolstering defense intelligence for a safer world

Vannevar Labs, a defense-tech startup supporting U.S. Department of Defense mission users, needed a better way to classify sentiment in news, blogs, and social media across multiple languages. Before working with Databricks, the team relied on GPT-4 and prompt engineering, but accuracy topped out around 65%, costs were too high, and lower-resourced languages like Tagalog proved especially difficult.

Using Databricks Mosaic AI, Vannevar Labs built a compound AI system and fine-tuned a domain-specific sentiment analysis model. Databricks helped them deploy it in just two weeks, cut latency by 75%, and achieve significant cost savings, enabling the team to scale more efficient AI-driven insights for mission-critical defense applications.

Vannevar Labs

Cane Punma

Senior Machine Learning Engineer