SARVESH GANESAN

Lead AI Architect

Architecting enterprise AI platforms that turn data into decisions

Get In Touch View Projects Download Resume

AI/ML

Engineering

Data

Engineering

Cloud

Infrastructure

About Me

Lead AI Architect with 2.5+ years building and shipping a cloud-native enterprise data platform end-to-end — from the agentic chat layer down to the Iceberg lakehouse and Kubernetes runtime images.

Solely architected 12+ production microservices on AWS EKS spanning agentic AI, distributed ETL, data lakehousing, MLOps, and containerized ML runtimes. Built a multi-agent orchestration system with 50+ dynamically-loaded data connectors that collapses hours of manual analytics into seconds of natural-language conversation. Delivered sub-second hybrid RAG at production scale and sub-10s queries over 100M+ row datasets via Spark / Iceberg / DuckDB.

Independently shipped a full-stack CRM platform for an enterprise tea distribution business — storefront, admin dashboard, invoicing, logistics KPIs, and a Claude-powered MCP server exposing 46 administrative tools across 9 business domains.

Multi-Agent AI

8 specialized agents orchestrated by Claude, 60+ tools, hybrid RAG with neural reranking, and cost-optimized model routing

Cloud & Infrastructure

12+ microservices on AWS EKS with Karpenter auto-scaling, 99.9% uptime, and multi-tenant isolation

Data & MLOps

TB-scale Iceberg lakehouse, 69+ data connectors, Spark/DuckDB runtimes, MLflow, and GPU training pipelines

Experience

09/2023 - 03/2026

Lead AI Architect

Groundzero Software Private Limited · Chennai

  • Solely architected and shipped 12+ production microservices end-to-end on AWS EKS — agentic chat platform, data provider API, ETL orchestrator, Iceberg REST catalog, MLOps control plane, and Spark/DuckDB/PyTorch/TensorFlow/scikit-learn runtime images
  • Built a Claude-orchestrated multi-agent chat system with intent-aware agent routing, per-tenant tool selection, and SSE-based streaming — enabling enterprise users to query databases, build dashboards, submit ETL jobs, and manage notebooks entirely through natural language
  • Engineered a dynamically-loaded connector architecture scaling data access to 50+ sources (Snowflake, BigQuery, Redshift, Postgres, MongoDB, Kafka, S3, Salesforce, MySQL, Oracle) with role-based authorization and schema introspection caching
  • Reduced RAG query latency by 90% (10s → sub-1s) using hybrid vector + full-text search with Cohere neural reranking, running over pgvector on production traffic
  • Delivered sub-10s analytic queries on 100M+ row datasets by architecting a TB-scale Apache Iceberg lakehouse backed by a gRPC-based REST catalog with schema evolution, time-travel, and multi-warehouse federation
  • Cut LLM inference costs via three-tier model routing (Opus/Sonnet/Haiku) with Anthropic prompt caching, per-request token tracking, and automatic downgrade for deterministic tool calls
  • Designed the ETL orchestration layer handling job submission, multi-job chaining, container log streaming from EKS pods, and dataset lineage — across three pluggable compute engines on Karpenter-provisioned spot nodes
  • Stood up the MLOps control plane from scratch — MLflow with custom JWT auth, Jupyter lifecycle management on EKS, fine-tune job dispatch across 7+ LLM providers, and GPU training containers for PyTorch, TensorFlow, and scikit-learn
  • Achieved 99.9% uptime across multi-tenant deployments via Karpenter just-in-time node provisioning, graceful pod lifecycle handling, and tenant-scoped compute isolation
  • Authored an end-to-end pytest suite covering dashboard rendering, visualization generation, and agentic chat flows with browser automation and screenshot-based regression validation
06/2023 - 08/2023

Software Trainee

SCI-BI Software Solutions Private Limited · Chennai

  • Developed and maintained data visualizations and dashboards using Power BI and Tableau
  • Participated in client meetings to gather requirements and translate them into technical specifications
  • Created reports and presentations, enhancing clients' data-driven decision-making

Technical Arsenal

AI & LLM

  • AWS Bedrock
  • Strands Agents
  • LangChain
  • RAG Systems
  • Cohere Embed & Rerank
  • Prompt Caching
  • Multi-Agent Orchestration

Cloud & Infrastructure

  • AWS EKS
  • S3
  • EC2
  • ECR
  • Lambda
  • CloudWatch
  • Docker
  • Kubernetes
  • Karpenter

Data Engineering

  • Apache Spark
  • Apache Iceberg
  • DuckDB
  • PostgreSQL
  • pgvector
  • Redis
  • Pinecone
  • MongoDB

Development & MLOps

  • Python
  • FastAPI
  • Flask
  • REST APIs
  • gRPC
  • SSE Streaming
  • MLflow
  • PyTorch
  • TensorFlow

Featured Projects

Sree Rajalakshmi CRM

End-to-end CRM and operations platform I built solo for a tea distribution business — customer storefront, admin dashboard, GST-compliant invoice generator, Razorpay checkout, WhatsApp order bot, and a Model Context Protocol (MCP) server exposing 46 administrative tools across 9 domains so Claude can run the business conversationally.

  • FastAPI
  • React
  • MCP Server
  • Razorpay
  • AWS Lambda
Visit Live Site

SRE Tea MCP Server

Production-grade Model Context Protocol server I shipped to let Claude Desktop, Claude Code, and custom MCP clients operate a live e-commerce business. Exposes 46 tools across orders, products, customers, inventory, invoicing, logistics, churn analytics, and warehouse management with automatic token refresh, parallel aggregation, and structured error recovery.

  • MCP
  • Node.js
  • Claude
  • STDIO Transport
  • JWT
View on GitHub

PlannerAgent

Conversational planning agent I built on LangGraph that handles 10+ turn dialogues without context loss, asks targeted clarifying questions, and versions plans across sessions. Ships with 12+ templates, interactive Gantt charts, multi-format import (Trello, CSV, Markdown), and a dual-provider LLM backend across Bedrock and Anthropic.

  • LangGraph
  • Claude
  • AWS Bedrock
  • Python
  • Session State
View on GitHub

QualiPilot

Production-grade data quality CLI and Python library for structural and statistical checks across CSV, Parquet, JSON, Pandas, Polars, Dask, and cuDF datasets. Generates deterministic JSON, HTML, and Markdown reports with CI severity gates plus optional LLM narration through AWS Bedrock, Ollama, or OpenAI-compatible endpoints.

  • Polars
  • Dask
  • cuDF
  • AWS Bedrock
  • Pydantic
View on GitHub

Tech Stack

  • Python
  • AWS Bedrock
  • Docker
  • PostgreSQL
  • Strands Agents
  • Apache Spark
  • Kubernetes
  • FastAPI
  • MLflow
  • Iceberg

GitHub Activity

--

Repositories

--

Total Stars

--

Recent Commits

--

Day Streak

Recent Activity

Loading activity...

Top Repositories

Loading repositories...

Let's Connect

Looking for an AI architect who ships production systems, not prototypes?

sarveshganesan2002@gmail.com
+91 8668063705
LinkedIn Profile
GitHub Profile