Ganesh Yallankar
Senior engineer building the tools and context layer AI agents run on.
Summary
Senior Member of Technical Staff at Salesforce on the Apex Platform team, with 6+ years in industry. Before Salesforce, Sprinklr. I focus on the infrastructure that makes AI agents useful in practice: accurate context, well-defined tools and a runtime you can trust.
Looking for: Senior or staff roles in AI agent engineering, agent platforms, or as a founding engineer.
Experience
-
Senior Member of Technical Staff
Member of Technical Staff, Jan 2022 to Jan 2026
- Led a type-information API that serves Apex code metadata to AI code-generation and developer tools.
- Brought Apex metadata into the metadata catalog and grounding service, so LLM-powered code features get accurate context through MCP.
- Moved AI code-generation and test-generation skills onto the new API and removed their dependency on an external MCP server.
- Profiled and diagnosed a tail-latency problem in compilation workloads; added lock-contention telemetry, SLO dashboards and regression alerts.
- Distributed-cache anomaly alerts, on-call, root-cause playbooks, feature-gate cleanups and backports.
- Technical subject-matter expert: scoped multi-sprint epics with product and scrum; mentored junior engineers; ran team demos and workshops that drove adoption of AI-assisted engineering.
-
Software Engineer
- Moved 500+ microservices from VMs to Kubernetes.
- Built a MongoDB metrics exporter (Prometheus, Kafka, Grafana on Kubernetes) and resource-utilization tooling that cut monthly cloud cost by about 20%.
- Automated disaster recovery for multi-service environments across 2,000+ database servers, with a team of ten.
- Built a monitoring and log-analysis platform in Java and Spring Boot, with parallel job scheduling on Apache Mesos and ZooKeeper, and Kafka and Elasticsearch data pipelines.
Projects
Repositories are private for now; these are write-ups. Full write-ups.
-
pi-mcp
Python, MCP, OAuth
An MCP server that gives different AI assistants one shared, durable memory. Per-path access rules, versioned writes with conflict checks, and human approval for proposed changes.
-
the-wardrobe
FastAPI, Postgres, React, MCP
A self-hosted app with a web UI for people and an MCP interface for agents. Presigned photo uploads instead of passing image bytes through the model.
-
Multi-agent trading research
Agents, research
Specialised agents research securities and report back to a coordinating layer. High level only; not open source.