~ › gokul › README.md

Gokul Rajendiran.

I build the platforms AI runs on — and the tools it uses.

4+ years engineering Zoho CRM's core — a 17,000-class multi-tenant platform. Along the way: a 128-tool MCP server and a dual-engine browser built for AI agents.

##Start here

• projects/crm-mcp-server.md128-tool MCP server
• projects/ai-browser.mda browser for AI agents
• skills.jsonthe full stack
• contact.mdsay hello

~ › gokul › about.md

About Me

Software Developer with 4+ years of enterprise CRM experience at Zoho Corporation, with a proven track record in building scalable backend systems.

With a B.E. in Computer Science from RMK College of Engineering (CGPA: 8.5, First Class with Honors), I bring a strong foundation in data structures, algorithms, and system design to every project I work on.

At Zoho, I've enhanced product quality for Zoho CRM, built and optimized complex database queries using Java and PostgreSQL — achieving up to 30% improvement in query performance — and led modernization initiatives contributing to product stability and faster release cycles. Oracle Cloud and Google Cloud certified, I thrive in cross-functional teams and love solving complex engineering challenges.

Beyond work, I'm an active competitive programmer with 5000+ problems solved across platforms, and I've been recognized in national hackathons including Smart India Hackathon.

##By the numbers

##What I do

~ › gokul › experience.md

Work Experience

~ › gokul › skills.json (rendered preview)

Technical Skills

##Core expertise — domains I own end-to-end

##Stack

##Professional skills

~ › gokul › projects › crm-mcp-server.md

python · fastmcp · stdio / sse / streamable-http

CRM MCP Server

A Model Context Protocol server that gives AI agents deep access to Zoho CRM's million-line codebase — with a self-learning usage layer, a knowledge graph, and a Java AST index covering 17,000 classes. The kind of tool this very portfolio's terminal is pretending to be.

##Highlights

• 128 tools across 28 groups — code search, ORM/DS-Query helpers, metadata lookups, and knowledge-graph queries, served over STDIO, SSE, and streamable HTTP so any MCP client can connect.
• Java AST index — 17,165 classes and 233,000+ methods parsed from CRM source, so agents jump straight to a definition instead of grepping a million lines.
• Self-learning layer — an aiosqlite usage store tracks which tools actually help and ranks them accordingly; the server gets sharper the more it's used.
• Knowledge graph — entities and relations extracted from the codebase connect modules, tables, and APIs so agents can reason about how pieces fit together.
• Built-in dashboard — a FastAPI + Jinja2 + HTMX web UI for inspecting tool usage and the learning layer in real time.

├─ code-index        java ast · 17k classes
├─ orm / ds-query    sql dsl bridge
├─ knowledge-graph   entities · relations
├─ usage-learning    self-ranking layer
└─ … 24 more groups
✓ 128 tools registered

~ › gokul › projects › ai-browser.md

typescript · lightpanda ⇄ chromium/cdp · manifest v3

AI Browser + claude-web-bridge

A dual-engine browser built for AI agents — Lightpanda for speed, Chromium over CDP for fidelity — plus a Manifest V3 bridge extension that lets Claude drive real, logged-in web apps: shadow DOM, accessibility tree, heavy SPAs and all.

##Highlights

• Two engines, one API — Lightpanda for fast, lightweight page work; Chromium via the DevTools Protocol when full rendering fidelity matters. The agent never has to care which one is underneath.
• Agent-grade actions — click-by-text, shadow-DOM-aware form filling, drag-and-drop, keyboard input, accessibility-tree snapshots, and network-idle waits — the primitives an AI needs to use real web apps reliably.
• claude-web-bridge — a Manifest V3 extension that drives the user's own logged-in browser through chrome.debugger: trusted clicks via Input.dispatchMouseEvent, JS via Runtime.evaluate, screenshots of background tabs.
• Shipped as a platform — a pnpm monorepo with an MCP server, a JS SDK, and a published Python SDK (ai-browser-sdk, Python 3.10–3.13), tested with vitest in CI.

window https://app.example
├─ navigation "Modules"
├─ button "New Lead"
└─ grid · 42 rows
> click(button "New Lead")
✓ done · via cdp session

~ › gokul › projects › zoho-crm-core.md

java · struts2 / spring · postgresql · redis · kafka

Zoho CRM Core Platform

The day job: the metadata and query internals of Zoho CRM — a 17,000-class multi-tenant SaaS. I work on the crmmeta query engine and module customization, tuning hot paths with a custom ORM, DS Query DSL, and multi-tier Redis caching.

##What I actually touch

• crmmeta query engine — the metadata layer that turns module and field definitions into SQL, with every org isolated by tenant in a single multi-tenant deployment.
• AdventNet persistence + DS Query — the in-house ORM and SQL DSL used across 5,000+ files of the codebase; I work on the query paths everything else sits on.
• Caching & performance — multi-tier caching (in-process + Redis) and query optimization that cut key response times by up to 30% in production.
• Platform plumbing at scale — Struts2 routing across 18 config files covering API v2–v10, Spring MVC with AOP interceptors, Kafka events, gRPC services, and Hystrix circuit breakers.

ds-query → sql
SELECT fields FROM CRM_FIELDS
  ⋈ CRM_MODULES (tenant)
cache L1 in-proc → L2 redis
✓ hot paths −30% response time

~ › gokul › projects › product-management.md

java · mvc · internal client framework · zoho internship (mar–jun 2022)

Product Management App

The internship project: picked up Zoho's internal client framework from scratch and built a complete product management application on Java MVC — recommendations, ratings, and quoting included. Built during the internship that converted into the full-time role.

##Highlights

• Internal framework, learned cold — ramped up on an in-house client framework with zero prior exposure and shipped a working application on it within the internship window.
• Product suggestions — a recommendation module that ranks products by a blend of user ratings and price.
• A ratings feedback loop — product ratings improve as users purchase and review, so the catalog gets more trustworthy the more it's used.
• Quotes & dynamic discounts — quote generation tied to user profile ratings: stronger profiles unlock better discounts.
• Requirements → features — translated business requirements into REST-aligned backend modules: recommendations, the discount engine, and the rating system.

> suggest(user)
rank: rating ⋈ price
purchase → review → rating ↑
> quote(user)
profile ↑ → discount ↑
✓ modules: suggest · rate · quote

~ › gokul › projects › earlier-work.md

Earlier Work

Hackathons & college projects.

~ › gokul › achievements.md

Achievements & Certifications

##Coding profiles

~ › gokul › education.md

Education

~ › gokul › contact.md

Get In Touch

I'm always open to discussing new opportunities, interesting projects, or just having a great conversation about tech. Feel free to reach out!