LLM Orchestration
LangChain & LangGraph workflows; conditional multi-agent execution without LLM overhead.
available for new AI builds
Devaraj
Kudumula
AI/ML Engineer & Researcher. Shipping production AI while researching NLP, agentic AI, LLMs, and computer vision.
~/devaraj — zsh
devaraj@ai:~/agents$ ask devaraj to
✓ LLMs · retrieval · agentic AI · evals · prompt eng
● LIVE
uptime 99.9%v4.7
3+
Years Experience
AI/ML
Engineer
LangGraph
& LangChain
system: online2026
LangChainLangGraphOpenAIGeminiPyTorchTensorFlowFastAPIMongoDBVector SearchDockerKubernetesGCPRAGAgentic AIReACTSelf-Healing AgentsLiteLLMHugging FaceLangChainLangGraphOpenAIGeminiPyTorchTensorFlowFastAPIMongoDBVector SearchDockerKubernetesGCPRAGAgentic AIReACTSelf-Healing AgentsLiteLLMHugging Face
01ABOUT
An engineer who ships AI. A researcher who keeps exploring.
I build enterprise-grade AI by day and dig into the research that powers it by night — LLM orchestration (LangChain, LangGraph), advanced RAG, and agentic frameworks, with active reading and tinkering across NLP, agentic AI, LLMs, and computer vision.
3+
Years Experience
AI/ML
Engineer
LangGraph
& LangChain
LLM
Orchestration
RAG
& Agentic AI
GenAI
In Production
// what I build
Advanced RAG
Query planner, self-query, agentic routing, multi-query rewriting, follow-ups for precision and coverage.
Agentic AI (ReACT)
Adaptive reasoning, self-healing tool use, and plug-and-play agent/tool integration.
Production ML
End-to-end lifecycle on GCP — FastAPI, MongoDB, Docker, Kubernetes, Jenkins, GitHub Actions.
// what I research
NLP
Natural Language Processing
Language understanding, transformers, semantic search, summarization, and dialog systems.
LLMs
Large Language Models
Architecture, fine-tuning, alignment, reasoning patterns, and prompt design.
Agents
Agentic AI
Multi-agent systems, ReACT reasoning, tool use, planning, and self-healing workflows.
CV
Computer Vision
Visual understanding, multimodal models, document AI, and image-language fusion.
02EXPERIENCE
From research prototype to production scale.
Three years building AI that ships, hardens, and earns trust at enterprise scale.
AI/ML Engineer
Intellect Design ArenaChennai, Tamil Nadu
- Architected and deployed enterprise LLM applications on Google Gemini & OpenAI via LangChain, serving 10,000+ daily active users across multiple business units.
- Spearheaded Enterprise GPT — a corporate-grade assistant with secure document Q&A, real-time web search, and a user-configurable agent marketplace.
- Designed a LangGraph orchestration system: conditional multi-agent workflows executed without LLM overhead — ~40% cost and ~50% latency reduction.
- Built advanced RAG: query planner, metadata-driven self-query filtering, Agentic RAG routing, and follow-up question loops for multi-entity coverage.
- Shipped a scalable Agentic AI (ReACT) framework with adaptive reasoning and self-healing tool recovery for robust multi-step agents.
- Built OCR + Chunk services on Azure Doc Intelligence with page/block/word metadata for high-fidelity document recreation and downstream AI ingestion.
LangChainLangGraphGeminiOpenAIFastAPIMongoDBDockerKubernetesGCP
Data Science Intern
Caterpillar Inc.Chennai, Tamil Nadu
- Built a Transformer-based NLP pipeline processing 100k+ comments/month with 92% sentiment classification accuracy and predictive trend insights.
- Fine-tuned a domain-specific Transformer summarizer — cut manual review time by ~90%, turning days into minutes.
- Designed a real-time analytics dashboard for service diagnostics and anomaly detection across distributed systems.
- Migrated legacy SAS pipelines to Python — better scalability, faster execution, lower infra cost.
TransformersPyTorchPythonPlotlySAS→Python
03PROJECTS
Things I've built. Things people use.
A selection of LLM systems, agentic architectures, and ML platforms — most in production today.
~/projects/enterprise-gpt
01Production · 10K+ DAU
Project 01: Enterprise GPT
Corporate-grade conversational assistant with secure document Q&A, real-time web search, and a user-configurable agent marketplace.
- Self-healing ReACT agents with adaptive reasoning
- Dynamic tool & agent selection based on user intent
- Scalable runtime serving 10K+ daily active users
LangChainLangGraphGeminiOpenAIFastAPIMongoDB
Impact
Faster knowledge retrievalLower inference costHigher decision velocity
01 / 06
04SKILLS
A pragmatic AI stack — tuned for reliability.
The tools and patterns I reach for to ship LLM systems that don't break in production.
Python
95%
LangChain
95%
OpenAI
92%
Git
92%
FastAPI
90%
PyTorch
88%
Docker
88%
Hugging Face
88%
MongoDB
85%
Gemini
85%
AWS
85%
Azure
85%
GCP
85%
GitHub Actions
85%
Bash
82%
TensorFlow
82%
Kubernetes
80%
Jenkins
80%
Plotly
75%
JavaScript
72%
// also fluent in — tap to expand
05RECOGNITION
Recognized work, certified craft.
A few highlights from the journey so far.
Spotlight Award×2
Intellect Design Arena
Two quarterly Spotlight Awards for high-impact enterprise AI delivery — Enterprise GPT and the RAG pipeline.
Google Data Analytics
Coursera
Professional Certificate in Data Analytics.
ML & Deep Learning Specialization
Udemy
End-to-end ML and deep learning specialization.
// education
Amrita Vishwa Vidyapeetham, Coimbatore
B.Tech. — Computer Science & Engineering (AI Specialization)
06CONTACT
Have an AI idea?
Let's build it together.
Open to roles, collaborations, and AI builds — RAG systems, agent platforms, LLM optimization, or anything where models meet real users.