Poorna Chandra Raju

Backend & AI Systems Engineer

Poorna Chandra Raju

I build high-performance backend infrastructure and production-grade AI systems — LLM deployment, RAG pipelines, and distributed systems for air-gapped, resource-constrained environments.

Tirupati, India mail@spcraju.com LinkedIn

01 / About

Backend engineer with 8+ years at Zoho Corporation building high-performance backend infrastructure for on-premise applications and production-grade AI systems. I have deep experience writing highly scalable enterprise web applications and optimizing backend design at multiple levels for high throughput — both for normal consumer use and for high-profile, high-secure environments like banks, governments, and militaries.

I specialize in LLM deployment, agentic pipelines, and RAG systems, with particular focus on distributed systems, MCP servers, document ingestion pipelines, and deploying AI solutions in air-gapped and resource-constrained environments.

02 / Skills

Programming Languages

Java, Python, C, SQL, Bash

Backend & AI

REST APIs, Tomcat, vLLM, MLX, RAG Systems, Agentic Pipelines, Standard AI Endpoints, MCPs, Docker

Database Systems

PostgreSQL, low-level UDFs, pgvector, Redis, Columnar Storage, DB Extensions

System Design

Distributed Systems, Job Queues, Async Processing, Caching, Fault Tolerance, Microservices

Performance Engineering

High-throughput Systems, Resource Optimization

GPU Computing

LLM deployment on NVIDIA GPU clusters, Tensor Parallelism, Semantic Embeddings, Hybrid Retrieval

03 / Projects

Selected work at Zoho Corporation as Senior Software Engineer (Dec 2017 – Present).

AI-Powered Automated Bug Fixing Agent

Jan 2026 – Present
  • Built a comprehensive RAG-based AI agent that automatically diagnoses and fixes bugs across Java, Python, JavaScript, and XML codebases using Tree-sitter AST analysis.
  • Implemented PostgreSQL with pgvector for hybrid vector and keyword retrieval, storing semantic embeddings for efficient code search.
  • Designed a multi-stage enrichment pipeline with AI-generated documentation, cross-encoder reranking, and structured LLM prompting for precise code-patch generation.
  • Developed agentic workflows with tool-calling loops, two-pass analysis, iterative fix-and-test cycles, and JVM debugger integration for live runtime inspection via MCPs.
  • Built an extensible architecture with pluggable LLM backends supporting OpenAI-compatible and Anthropic APIs, dependency-graph traversal, and a domain-specific knowledge base.
RAGpgvectorTree-sitterMCPAgentic

Local LLM Deployment Infrastructure

Dec 2025 – Present
  • Successfully deployed LLMs up to 120B parameters on local NVIDIA GPUs using a vLLM wrapper for local coding assistance.
  • Developed a Python wrapper on MLX for Mac Studio deployment and inference optimization in air-gapped environments.
  • Implemented tensor parallelism and advanced caching, achieving 300% throughput improvements with limited hardware resources.
  • Enabled developer teams to access local LLM capabilities, cutting down external AI dependencies.
vLLMMLXTensor ParallelismAir-gapped

Mock Data Generation Using LLM & Data Generator Engine

Sep 2025 – Nov 2025
  • Replaced manual JSON schema creation with LLM-generated schemas based on natural-language criteria using function-calling.
  • Built an automated validation pipeline with static validation, job-pool submission, and polling mechanisms for enterprise-scale data generation.
  • Generated correlated multi-table datasets with primary and secondary key relationships, producing millions of rows for realistic testing scenarios.
  • Integrated with automation pipelines for high-volume data requirements, reducing data-prep time from hours to seconds.
LLMFunction-callingJob Queues

Product Management Agent Using LLM

May 2025 – Sep 2025
  • Built a comprehensive agentic pipeline to assist product-management teams using Crew AI and LangChain frameworks.
  • Designed automated data-collection and analysis systems for connector integrations and document ingestion.
  • Implemented intelligent reporting with natural-language to structured-output conversion.
Crew AILangChainAgentic

Data-Specific Report Generation Pipeline

Oct 2024 – Apr 2025
  • Designed and implemented an end-to-end LLM agentic pipeline for automated report generation using LangChain.
  • Built SQL and Python code-generation capabilities for complex data-manipulation tasks.
  • Deployed the pipeline for PM-team usage to enhance existing integration templates for tens of integrations.
LangChainCode GenerationSQL

Python ML Functions in Postgres

Oct 2021 – Sep 2024
  • Developed high-performance Python ML functions wrapped as PostgreSQL user-defined functions for real-time inference.
  • Designed for cloud application usage with native-level performance optimizations achieving sub-millisecond response times.
  • Integrated ML capabilities directly into the database layer, enabling real-time AI features for millions of users.
PostgreSQLUDFsML Inference

Mock Data Generator for High-Load Testing

Apr 2020 – Sep 2021
  • Built a comprehensive mock-data-generation web application (Mockaroo-style) for enterprise testing.
  • Implemented flexible schema-based data generation for performance testing under high-load conditions.
  • Enabled system-performance evaluation across multiple Zoho products supporting 1M+ concurrent users.
Web AppLoad TestingSchema-based

Columnar Storage Database Engine

Dec 2018 – Mar 2020
  • Ported a Linux-based PostgreSQL columnar-storage extension to Windows for high-performance analytic queries.
  • Optimized for enterprise environments with 10x performance improvements for analytical workloads.
  • Ensured seamless integration with existing Zoho infrastructure, supporting terabyte-scale data processing.
PostgreSQLColumnar StorageC

Server Backend Performance Engineering

Dec 2017 – Nov 2018
  • Identified and resolved critical bottlenecks in resource-constrained server environments.
  • Optimized systems for heavy task loads with minimal resource allocation, improving performance by 70%.
  • Improved system reliability and performance across multiple Zoho services serving millions of users.
PerformanceOptimizationDistributed

04 / Education

B.Tech — Electronics & Communication Engineering

2017

S. V. College of Engineering, Tirupati

JNTU, Anantapur

Higher Secondary

2013

Sri Gayatri Junior College, Tirupati

Intermediate Board of Education, Andhra Pradesh

Secondary School

2011

Sri Vignana Sudha High School, Tirupati

Board of Secondary Education, Andhra Pradesh