Pixel-art portrait of Chitraksh: curly hair, sunglasses, cream kurta, cargo pants, backpack over one shoulder

I build AI agents that query your data, talk back by voice and hold up in production.

See my workSay hi
Agentic AI
LLMs
LangGraph
RAG
MCP
vLLM
Vector DBs
Voice agents
Text-to-SQL
FastAPI
Machine learning
PyTorch

hi, I'm
Chitraksh

I like the part of AI that has to survive production: retrieval that is actually relevant, SQL that is actually valid, a voice that answers before you lose patience.

I started as a frontend intern, then moved into AI: voice-driven workflow automation and AI avatars, then lead generation and data enrichment pipelines. Today I build AI projects end to end: multi-agent systems in LangGraph, hybrid retrieval, real-time voice, and the backends around GPT and other LLMs, with guardrails and tracing to keep them reliable. I even built a GPT from scratch to see what happens inside one.

  • Now

    Software engineer at Prismberry, building AgentIQ

  • Studied

    B.Tech in Computer Science, ABES Engineering College, 2022 to 2026

  • Research

    IEEE SCEECS 2026 paper on emotion-aware AI agents

  • Freelance

    Rising Talent on Upwork: AI agents, RAG pipelines and voice AI for clients

selected work

Newest first. Most of these run live, and every one has its code on GitHub.

AgentIQ

Prismberry Technologies, Feb 2026 to now

An enterprise agent platform that lets people query production databases in plain language. A 9-stage text-to-SQL pipeline combines BM25, vector and foreign-key-graph retrieval, plans in two LLM stages and validates every query with sqlglot. A voice agent answers the same questions out loud.

Company project, so there is no public repo or demo.

  • Sarvam STT + OpenAI Realtime voice agent
  • LiteLLM multi-provider routing on a LangGraph state machine
  • Hybrid RAG with Cohere reranking, SSE streaming, Celery ingestion
  • PII guardrails with Microsoft Presidio
GPT Anatomy screenshot

GPT Anatomy

Oct 2026

A 10.8M-parameter GPT written from scratch in PyTorch, plus a 3D site that opens it up. Every attention weight, neuron and probability on screen comes from the trained model. Thread tuning made generation 8× faster on a throttled CPU.

  • PyTorch
  • FastAPI
  • SSE
  • Three.js
  • Docker
Live demo Code
MeetMind screenshot

MeetMind

Jun 2026

A meeting bot that joins Meet or Zoom, transcribes, and drafts Jira tickets with LLM self-correction. A 7-node LangGraph agent pauses for human review before anything is pushed, and survives restarts through Postgres checkpoints.

  • LangGraph
  • FastAPI
  • React
  • PostgreSQL
  • GPT-4o
Live demo Code

Agentic Blog Writer

May 2026

Researches a topic, plans the article, writes every section in parallel with evidence grounding, then generates the diagrams. A router picks closed-book, hybrid or open-book research per topic.

  • LangGraph
  • Tavily
  • Pydantic
  • Gemini
Code
Lumira screenshot

Lumira

May 2026

An AI therapy companion and my final-year project: voice conversations with an empathetic therapist, session insights and secure accounts. It grew into the IEEE paper below.

  • Next.js 15
  • Vapi
  • Gemini
  • Firebase
Live demo Code

Production AI Support Agent

Jan 2026

Drop in PDFs and ask questions. Ingestion, chunking, embeddings and retrieval are orchestrated as Inngest functions over a Qdrant vector store.

  • RAG
  • Qdrant
  • Inngest
  • Streamlit
Live demo Code
Medical Chatbot: Vision & Voice screenshot

Medical Chatbot: Vision & Voice

Dec 2025

Speak your symptoms, upload a photo, get an answer read back to you in under ten seconds. LLaMA-4 Scout for vision, Whisper for speech, ElevenLabs for the voice.

  • LLaMA-4
  • Whisper
  • ElevenLabs
  • Gradio
Live demo Code

Multi-Source RAG Agent

Jul 2025

A LangChain agent that decides whether to answer from Wikipedia, Arxiv or an indexed documentation set, and cites where the answer came from.

  • LangChain
  • FAISS
  • OpenAI
  • Arxiv
Code

AI Avatar Agent

Jun 2025

A conversational agent with a face: it joins a LiveKit room, listens, thinks with OpenAI and answers through a Tavus avatar in real time.

  • LiveKit
  • Tavus
  • OpenAI
  • Deepgram
Code
All repositories on GitHub

where I've
shipped

  1. Prismberry Technologies

    Software Engineer, AI / Backend

    Feb 2026 to now · Noida

    • Built AgentIQ: a 9-stage text-to-SQL pipeline (hybrid BM25 + vector + FK-graph retrieval, two-stage LLM planning, sqlglot validation) and a voice agent for natural-language querying over production databases.
    • Engineered multi-provider LLM orchestration with LiteLLM and a LangGraph state machine, hybrid RAG with Cohere reranking, SSE streaming, Celery ingestion and PII guardrails via Microsoft Presidio.

  2. The Ladder

    AI Automation Engineer

    Jul 2025 to Jan 2026 · Gurugram, hybrid

    • Built automation and RAG pipelines in n8n (40+ node workflows) with Apollo, Clay and Apify for lead generation, HubSpot CRM enrichment and AI personalization for enterprise clients.
    • Developed inbound and outbound voice agents for sales and support, plus ad-creative pipelines (Midjourney, DALL·E, GPT) that cut manual effort and cost.

  3. Dreamable Inc.

    AI Agent Intern

    Jun to Aug 2025 · San Francisco Bay Area, remote

    • Built agents with n8n, LangChain and OpenAI for WhatsApp, LinkedIn and lead workflows.
    • Designed RAG document Q&A over embeddings and vector databases; engineered voice agents with Whisper, ElevenLabs and Tavus.

toolbox

Languages

  • Python
  • C / C++
  • SQL

AI & ML

  • Agentic AI
  • Generative AI
  • Machine learning
  • Deep learning
  • NLP
  • Multimodal AI
  • Voice & speech AI
  • Text-to-SQL
  • Prompt engineering

LLMs & models

  • OpenAI GPT-4o
  • Gemini
  • LLaMA-4
  • Groq
  • vLLM
  • Whisper
  • Sarvam STT
  • ElevenLabs
  • Tavus

Vector DBs & retrieval

  • FAISS
  • Milvus
  • Qdrant
  • Pinecone
  • BM25
  • Hybrid search

AI frameworks

  • PyTorch
  • LangChain
  • LangGraph
  • LiteLLM
  • LlamaIndex
  • LangSmith
  • MCP
  • LiveKit Agents
  • Pipecat
  • Vapi
  • Inngest
  • Cohere rerank
  • Multi-agent
  • RAG

Automation & tools

  • n8n
  • Relevance AI
  • Apify
  • Git
  • Linux

Backend & infra

  • FastAPI
  • PostgreSQL
  • Supabase
  • Celery
  • SSE streaming
  • Docker
  • Presidio
  • sqlglot

emotion-aware agents

Advances in Emotion-Aware AI Agents for Mental & Emotional Health

A multimodal agent that reads text, speech and facial emotion together, to support people through hard conversations. Published at IEEE SCEECS 2026.

Read the paper ↗

IEEE SCEECS
2026

hire me

I take on freelance projects through Upwork or direct. Here is what I have done for clients, and what I can do for you.

AI & voice agents

Agents that book, answer and escalate: chat, WhatsApp or phone, with a human in the loop where it matters.

RAG pipelines

Search and Q&A over your documents and databases, with retrieval that is actually relevant and answers that cite sources.

Automation

n8n and Python workflows for lead generation, CRM enrichment and the repetitive work your team does by hand.

Vibe-coded to production-ready

You built it with AI and it works on your laptop. I turn it into something you can put customers on.

Hire me on Upwork Email me
  • Auth, secrets and input validation
  • Error handling, retries and timeouts
  • Tests for the parts that must not break
  • Docker, CI and a real deploy
  • Logging, tracing and cost / latency checks
  • A short handover so your team can keep going
  • Rising Talent on Upwork
  • 5.0 on both rated jobs
  • 198 hours billed
  • Replies within 0 to 4 hours
  • 30+ hrs a week
  • Open to contract-to-hire
“It was very pleasant to work with Chitraksh. He is a very skilled developer, very good at communicating, and understood the problem and solved it quickly.”
Client review on Upwork, June 2026