FinchAI - Optimization

Overview

Led a comprehensive assessment of its Elasticsearch platform, which underpins its core product. The engagement focused on identifying performance bottlenecks, evaluating the impact of advanced query patterns (including k‑nearest neighbor queries used by LLM agents), and producing a strategic optimization roadmap to support Finch’s growth and scalability objectives.

The Challenge

After having initial success with its platform, Finch was experiencing significant performance challenges within its deployment as KNN searches, particularly those executed by Agentic AI were increasing. This was caused by:

• Complex queries and aggregations
• KNN searches supporting LLM‑driven workflows
• Salience ranking operations

Solution

Conducted a structured Elasticsearch performance assessment designed to establish both short‑term remediation guidance and a longer‑term optimization roadmap.

Key activities included:

• Current‑state performance investigation, analyzing query logs and usage patterns across all environments to identify bottlenecks affecting responsiveness and throughput.
• Query pattern and aggregation review, focusing on slow queries, large result sets, and inefficient filtering behavior
• KNN impact analysis, assessing how computationally intensive vector queries initiated by LLM agents were affecting overall cluster performance and concurrency
• Salience ranking optimization, exploring ways to improve the efficiency of salience ranking operations in Elasticsearch Findings review and recommendations, delivered via a written assessment report and discussed with the engineering team in working sessions.
• Assisted with Performance Testing, helped design and execute performance optimizations tested through sustained load testing.

The outcome of this phase was a documented optimization roadmap, providing clear guidance on addressing immediate performance issues while establishing a foundation for future scaling and enterprise readiness.

Technologies Used

• Elasticsearch

Results

“You’re going to go tell everyone that you solved our performance problems in two days.” — Engineering Team Member

• Delivered a comprehensive performance assessment of the Elasticsearch implementation, grounded in real query patterns and production usage
• Identified key contributors to performance degradation, including expensive aggregations and high‑cost KNN query workloads
• Provided a prioritized optimization roadmap to guide remediation efforts and inform future architectural decisions as the platform scales