AI-Native Full Stack Developer

Hi, I'm Vaibhav Pundir

Building agentic systems, automation workflows, and backend-heavy products that turn AI into reliable software.

Current Focus

AI-native products with strong backend foundations

Available

Agentic systems

Tool calling, memory, context, multi-step workflows

Automation engines

Queues, schedulers, integrations, background workers

Backend-heavy AI apps

APIs, auth, data pipelines, observability, reliability

Work Experience

My work across agentic systems, automations, and backend platforms

Software Development Engineer

Codemate.ai

2025 - 2026

Built and maintained the React-based webview UI (chat interface, task history, settings panels) using TypeScript, implementing real-time streaming of LLM responses via VS Code's message-passing API between the extension host and webview.
Designed and extended the core agent orchestration layer in Node.js/TypeScript, handling tool-call parsing (XML and native function-calling formats), multi-turn conversation state, and provider-agnostic integration across LLM APIs (Anthropic, OpenAI, OpenRouter, local models)
Implemented file system and terminal tool handlers (read/write/diff, shell execution, browser automation) with permission-gating and diff-preview UI, enabling safe autonomous code edits within the editor.
Optimized context management for large codebases by building token-aware truncation, sliding-window conversation history, and workspace file indexing to keep the agent within model context limits.
Contributed to the extension's build and packaging pipeline (esbuild, VS Code Marketplace publishing) and added telemetry/error-tracking to monitor tool-call success rates and reduce agent failure loops in production.

Projects

Projects I've built

Cobraine Telegram Bot

Cobraine transforms Telegram into a personal AI second brain for capturing ideas, remembering context, scheduling reminders, and getting intelligent assistance. Built as an agentic system with a backend-heavy architecture for real-world automation workflows.

NodeJSAgentic AILanggraph/LangchainPostgreSQLRedisBullMQDocker

Google Workspace MCP Server

A multi-tenant MCP server that gives AI agents secure access to Google Workspace across Drive, Gmail, Calendar, Tasks, Docs, and Sheets through a stateless HTTP interface with connection-level auth.

NodeJSTypeScriptMCPGoogle APIsOAuth 2.0HTTP

Skills & Technologies

AI, backend, and automation tools I use to build production systems

AI & Agents

LLM Tool CallingAgent OrchestrationLangGraphMCPVector Search

Backend Engineering

Node.jsTypeScriptGoExpressREST APIs

Automations

BullMQRedis QueuesScheduled JobsWorkflow AutomationAPI Integrations

Full Stack UI

ReactNext.jsTailwind CSSWebviews

Data Layer

PostgreSQLRedisMongoDBpgvector

Infrastructure

DockerKubernetesAWSGCPObservability