Geeta Shankar

About Geeta Shankar

I make large-scale systems fast, reliable, and ready to ship. At Salesforce I\'ve been the sole performance engineer behind some of the company\'s biggest bets — including the go/no-go launch decision for a feature that rolled out to customers globally, and scaling a developer platform 100× in six months by working directly with AWS to tune EC2 and warm pool configurations. What makes my work unusual is the combination: I sit at the intersection of classical performance engineering (latency profiling, distributed tracing, benchmarking, phased rollout validation) and modern AI systems (LLMs in production, semantic search, anomaly detection, RAG). A few things I\'ve shipped- 99%-accurate anomaly detection system that catches regressions before they reach customers - GPT-powered natural language coding feature serving developers - Performance architecture for a platform that scaled 100× without degrading latency I\'ve spoken at Dreamforce, ODSC (emceed the LLMs & GenAI track), Trino Summit, and PyLadies on topics ranging from production anomaly detection to performance engineering in Python. I\'m based in San Francisco and always interested in conversations about hard performance problems at the intersection of AI and infrastructure.

Experience

1. Senior Software Engineer

   Salesforce

   Jan 2022 — Present · San Francisco, CA, US

   SMTS 2026 – PresentPromoted to SMTS in 2026 for technical leadership and cross-team influence while concurrently owning two major enterprise-scale performance programs with no overlap.· Sole perf engineer for go/no-go decision on Security Center feature rolling out to customers — the largest scaling effort on Salesforce\'s yearly company vision. Cleared for full launch.· Scaled Agentforce Vibes 100× in 6 months via AWS EC2 and warm pool tuning — owned the performance architecture end to end.MTS 2023 – 2026Performance engineer across three product surfaces simultaneously: Agentforce Vibes, Security Center, and Digital Wallet.· Prevented regressions across developers with a 99%-accurate anomaly detection system that caught issues before customers saw them.· Reduced Apex Language Server cold-start latency 50% for developers.· Reduced task completion time 30% by building a GPT-powered natural language coding feature.· Unified metrics across disparate databases into a harmonized data model powering Security Center dashboards.AMTS Jan 2022 – 2023Joined as AMTS and promoted to MTS within 1.5 years. Built Splunk & Perfkit API pipelines for Apex workload analysis and ML models predicting workload issues with 96% accuracy, reducing detection time 30%.

Education

• University of California, Berkeley

  Data Science, (Computer Science. Machine Learning, Biology)