Skip to content
Kelavath Balaji Naik

Applied AI Engineer · Samco Securities · India

KelavathBalaji Naik

Agentic AI · RAG · MCP · Production LLM Systems

GenAI and Applied AI Engineer building and operating production AI systems across voice AI, RAG, LLM infrastructure, MCP tooling and multi-agent automation. I also write about applied AI engineering on LinkedIn.

Featured project: PromptForge
View my work
Get in touch

1+

Years in production AI

20+

Open-source projects

2

Invited guest lectures

1

IEEE publication

01 — Skills

What I work with

Agentic AI & LLM

MCPA2ALangGraphLangChainLiteLLMTool callingStructured outputsMulti-agent systemsOpenAIClaudeGemini

RAG & Retrieval

Hybrid search (dense + BM25)RRFLLM rerankingQdrantfastembedChunkingEmbeddingsVector databases

Languages

PythonTypeScriptJavaScriptSQL

Backend & Data

FastAPISQLAlchemyAlembicPydanticasyncioRESTGraphQLMySQLPostgreSQLMongoDBRedisClickHouse

LLMOps & DevOps

Streaming (SSE / WebSockets)Cost trackingOAuth 2.1 / JWTDockerKubernetesJenkinsNginxCI/CDGrafana

Voice & Frontend

LiveKit AgentsSIPSTT / TTS pipelinesSpeech-to-speechReactNext.jsTypeScriptReact FlowTailwind

Model landscape I track and use

Hosted models

OpenAI GPT
Claude
Google Gemini
xAI Grok

Speech models

OpenAI Realtime
Gemini Live
ElevenLabs
Grok Voice

Open-weight via Ollama / NVIDIA

Ollama
NVIDIA NIM
Qwen
Llama
Mistral
Gemma
DeepSeek

Protocols & routing

MCP
A2A
LiteLLM
OpenAI-compatible

02 — Selected Work

Selected work

All projects
01Developer Tools

PromptForge

Refines rough requests into clearer prompts inside browser chats and coding agents. Review the rewrite, see what changed, and use your own model provider.

TypeScript
Chrome MV3
MCP
LLMs
CodeInstall
PromptForge rewriting a request in ChatGPT, with explanations and coaching. Sample output from a test model.

Browser extension demo · Sample output from a test model

02Agentic AI

Agent2Agent — Multi-Agent Orchestration

Agent-to-agent communication, tool use, and autonomous task delegation between cooperating LLM agents.

Python
Agents
LLMs
Tool Use
Agent2Agent — Multi-Agent Orchestration
03Agentic AI

Voice Agent — Retell AI

Real-time conversational voice agent that handles phone-style interactions end-to-end on an LLM-voice platform.

TypeScript
Retell AI
Voice
LLM
Voice Agent — Retell AI
04LLM / RAG

Semantic Search Engine

Search by meaning, not keywords, with vector embeddings and Next.js. The foundation of a production RAG pipeline.

Next.js
Vector DB
Embeddings
RAG
Semantic Search Engine

03 — Writing

I write about applied AI

Content creator on LinkedIn

Regular posts on what it takes to run AI in production: agent design, RAG that holds up, LLM infrastructure, MCP tooling, and the open-source model landscape.

Follow on LinkedIn

What I write about

  • Agentic AI & MCP

    Tool design, multi-agent patterns, what breaks in production.

  • RAG that holds up

    Hybrid search, reranking, evaluation and answer grounding.

  • LLM infrastructure

    Gateways, cost control, streaming, provider fallback.

  • Model landscape

    New releases, open-weight models, running them with Ollama.

  • System design for AI

    Full-stack architecture around models, from API to UI.

Currently

Applied AI Engineer @ Samco Securities

Voice AI, RAG systems, LLM infrastructure and MCP tooling in production.