Jayakrishna Gandhamalla

Short on time? This assistant knows my résumé and answers questions about my work. Fittingly, it’s an LLM agent.

1. 2025 — 2026

   Xpeng MotorsSanta Clara, CA

   • Senior Software Engineer, AI PlatformSep 2026 — Nov 2026
   • Software Engineer, AI PlatformJul 2025 — Aug 2026

   Built Zod, Xpeng’s internal AI coding-agent platform, from zero to production: 1,000 users (600 daily) and 50 billion tokens per workday. Lead engineer and top contributor (about 5,700 of 9,600 commits) on a ~1M-line monorepo covering the agent runtime, eight backend services, and desktop, CLI, and VS Code clients.

   • Built the Python agent runtime: resumable streaming, crash and interruption recovery, context compaction, permission-aware subagents, Git worktree-isolated workspaces, and delegation to Claude and Codex over ACP.
   • Cut time to first token from 4.4s to under 0.6s with agent prewarming. Content-hash file editing and ripgrep-backed search cut model steps by 11–20% in A/B tests.
   • Built an OpenAI-compatible LLM gateway with provider routing, rate limits, and usage tracking. Differential request caching cut median upload time per agent step from ~18s to 0.03s.
   • Shipped an agent SDK for headless CI that automatically reviews GitLab merge requests once lint, tests, and type checks pass, plus a marketplace for skills, MCP servers, and agents.
   • Added agent safety and memory: tool approvals, an AST-based shell-command safety classifier, automatic review of risky calls, and persistent memory with BM25 search and nightly distillation.
   • Built Space Assistant, a persistent agent that spins up and steers worker chats, monitors outcomes, and writes daily recaps.
   • Packaged the runtime as a Docker-free single binary for six platforms with staged-rollout release pipelines. Cut the CI test gate from 49 to ~19 minutes and unchanged rebuilds from ~4 minutes to ~1 second with Turborepo.

   PythonFastAPIasyncioTypeScriptReact 19ElectronMCPACPpgvectorTurborepo

2. 2025

   Xpeng MotorsSanta Clara, CA

   • Tools & Infrastructure InternJan 2025 — Jun 2025

   Developer infrastructure for the autonomous-driving perception team, plus early agent tooling.

   • Migrated the perception codebase’s lint, type-check, and test CI to Bazel (700+ commits), with test sharding, remote execution, Hugging Face caches, daily coverage, and flaky-test retries.
   • Forked and customized Aspect’s rules_py, cutting integration-test runs from a couple of hours to under 20 minutes and freeing GPU capacity.
   • Built an A2A multi-agent framework with web, CLI, and VS Code interfaces and MLflow tracing and evaluation, adopted by 20+ teams. Also built an LLM unit-test generator with an LLM-graded evaluation tool.

   Bazelrules_pyPythonA2AMLflowGitLab CIMongoDB

3. 2020 — 2023

   Siemens EDAHyderabad, India

   • Software Engineer IIAug 2020 — Aug 2023

   Led full-stack development of Capital Teamwork and Xcelerator Start from concept to beta, deployed at clients including Lockheed Martin.

   • Built an infinite-canvas collaboration tool with real-time sync and a cross-platform launcher (Electron + Java) that connects web UIs to local executables.
   • Merged Mendix, React, and Angular components into one UI library, cutting front-end development time by 15%.
   • Set up GitLab CI/CD with automated testing, and migrated a large AngularJS app to Angular 14 in two days.

   TypeScriptAngularReactNode.jsElectronJavaGitLab CI

4. 2020

   Being ZeroHyderabad, India

   • Software EngineerMar 2020 — Jul 2020

   Built a MEAN-stack coding-contest platform with Monaco Editor and external (VJudge-style) submissions.

   MongoDBExpressAngularNode.js

• ChessVision2024

  Real-time chess commentary. YOLO and OpenCV turn a camera frame into a board position, Stockfish analyzes it, and an LLM explains the move. Fine-tuned detectors improved mAP by 12.6%.

  YOLODETROpenCVStockfishLLMs
• Chronoscribe↗2024

  Microservices for live lecture transcription and RAG summarization over SSE, shown at UB CSE Demo Days. Fine-tuned BERT summarization improved ROUGE-L by 11.7%.

  FastAPINext.jsRedisPineconeLangChain
• Vision Transformer from scratch2024

  A ViT image classifier implemented in PyTorch from first principles, reaching 85.4% accuracy (3.7 points above CNN baselines).

  PyTorch

AI & agents

Agent runtimes · Multi-agent orchestration · Tool calling · Context & memory · MCP · ACP · A2A · Agent SDKs · AI code review · LLM gateways · RAG · Evals (MLflow, GEPA) · OpenAI & Anthropic APIs · Amazon Bedrock · PyTorch

Languages

Python · TypeScript · JavaScript · Java · C++

Backend & infra

FastAPI · asyncio · Node.js · NestJS · WebSockets · SSE · Kong · PostgreSQL · pgvector · MongoDB · Redis · SQLite

Build & dev infra

Bazel · rules_py · Turborepo · GitLab CI/CD · Docker · Kubernetes · GCP · Pytest · Playwright · Jest

Clients

React 19 · Next.js · Electron · VS Code extensions · CLI tools · Tailwind · Angular

• Your Comprehensive Guide to SUNY Buffalo: Fall 2024 EditionJul 2024
• My Journey from Chaos to OrderNov 2023

Hiring for agent infrastructure, AI platform, or senior full-stack work?

I’m open to new roles in the San Francisco Bay Area. Email is the fastest way to reach me.