INSA Strasbourg powers new research database with ParadeDB. Read their story.

Sweetspot Unifies Hybrid Search on Postgres with ParadeDB

By Ming Ying on June 9, 2024

Sweetspot Overview

Sweetspot is an AI-driven platform for government procurement and contracting. Sweetspot streamlines the entire contracting process from opportunity identification to bid/no-bid decision making to proposal development.

The Challenge

Sweetspot needed a hybrid search — the combination of full-text and vector search — solution to power its contract search and AI-assisted proposal drafting products.

Since Sweetspot was already using Postgres, it was important that their search solution interoperated seamlessly with Postgres. Before using ParadeDB, Sweetspot experienced significant friction synchronizing large volumes of data between Postgres and external databases like Elasticsearch. This synchronization involved writing and maintaining custom connectors, which were prone to bugs and data inconsistencies. Additionally, many of the vector databases that Sweetspot evaluated had unreliable infrastructure which caused frequent downtime in their application.

The Solution

Sweetspot needed a search solution that was easy to integrate, compatible with existing Postgres architecture, and offered reliable performance over 15 million documents and 2 million vectors.

After an extensive review, Sweetspot selected ParadeDB. With ParadeDB, Sweetspot unified its data stack on Postgres, which removed the need to manage a separate database or use custom connectors. ParadeDB's advanced full-text and vector search capabilities enabled Sweetspot to build powerful hybrid search experiences without additional operational complexity.

Outcomes

Integrating with ParadeDB took less than a week since Sweetspot was already using Postgres and could re-use its existing code.

Post-integration, Sweetspot observed a 50% overall reduction in query latency. Hybrid search queries achieved a P99 latency of 50 milliseconds and had improved precision and recall compared to pure vector search.

ParadeDB also simplifies the process of achieving FedRAMP and NIST compliance since Postgres has robust authentication, access control, logging, and disaster recovery built in.

If you think that ParadeDB can solve a use case for your organization, we invite you to contact us.