AI
Engineer
& Builder
Applied AI Engineer
Open to roles AI Engineer @ Calyb AI LangGraph · RAG · Knowledge Graphs Savarkundla, IN, remote

I build the backend of AI products. Right now that means agent orchestration and a retrieval layer that turns raw user activity into workflows a machine can run.

Harsh Rathod
Savarkundla, India. Working remote.

Scroll ↓

LLM ApplicationsMulti-Agent SystemsRAGKnowledge GraphsOntology DesignWorkflow AutomationLangGraphNeo4jPython
LLM ApplicationsMulti-Agent SystemsRAGKnowledge GraphsOntology DesignWorkflow AutomationLangGraphNeo4jPython
The loop

This is the part everyone is building right now. A model that plans, acts, looks at what came back, and decides again.

Getting one pass to work is a demo. Getting the loop to stay correct over many turns, under a token budget, with tools that fail, is the job.

running plan 0 tokens
Context window · what actually fits 0 / 128,000 tokens
System prompt Retrieved context History Free
01 / About
Attention · queries vs keyst=0
0.01.0 · each row sums to 1

I work on the applied side of AI, shipping systems rather than training models. Most of my work sits underneath the product rather than in it: the orchestration layer, the retrieval layer, and the graph that holds what a system knows.

At Calyb AI I design a backend intelligence layer that models what users do in a UI as a graph ontology, then compiles it into workflows other systems can execute. It ships as an SDK with separate development and production modes. Before that I studied ICT at Dhirubhai Ambani University.

3Live signals fused in Trueyy
4Agent roles orchestrated
50+Tested user sessions
02 / Selected work
01

Trueyy Live site

Real-time AI-use detection for live remote interviews. I built the detection side: the signal pipeline that reads app focus, paste velocity, and the structural signature of an answer as it is typed or spoken, and the scoring loop that reweighs those signals every 30 seconds into one integrity score the interviewer watches live. Detection keys on output structure rather than keyword lists, so rewording a prompt does not defeat it — it carries recognition patterns for ChatGPT, Claude, Gemini, Copilot, Cluely, InterviewCoder and 50+ others. No signal is treated as proof on its own; every flag is timestamped so a human can go back and judge the moment themselves, which is what keeps false positives from turning into accusations. Runs browser-side beside Zoom, Meet, and Teams with a consent screen and no candidate install. trueyy.com
Live · 2026
02

PageIndex PR #314 & #383

Open source, VectifyAI/PageIndex (34k stars). PR #314: Made indexing idempotent by keying documents on canonical path so existing doc_ids are reused, eliminating UUID sprawl and orphaned JSON summaries. PR #383: Designed node-level text & summary versioning with deferred regeneration, cutting LLM re-indexing token costs by 92% on incremental edits. Read case study →
Open Source
03

Knowledge Compiler

End-to-end event-sourced knowledge compiler that digests unstructured enterprise data into a 100,000+ node Context Graph in Neo4j. 6-stage bitemporal pipeline, AST chunking, FastRP topological embeddings for instant alias deduplication, and a ReAct GraphRAG agent with zero hallucinations.
Private · 2026
04

ReFineAI

FastAPI backend streaming over SSE, Gemini 2.5 Flash and Groq behind one interface, with URL context tools. Ships with a C# desktop client and a separate database UI.
Private · 2026
05

Any Maker

LangGraph multi-agent pipeline. A planner breaks the prompt down, an architect turns it into file specs, and a coder reads the tree and edits only what changed. Human review between steps.
Private · 2025
06

AI Shopping Agent

Flask, LangChain, Groq. Product search and comparison over Tavily and SerpAPI, with currency conversion, behind a chat interface.
Private · 2025
07

Job Recommendation Engine

Flask, LangChain, Groq. Parses a PDF resume, generates job titles from it, pulls live listings through Google Jobs, then ranks them by cosine similarity over embeddings.
Private · 2025
08

WhatsApp AI Assistant

Twilio WhatsApp Business API, RAG, LangChain. Built during my internship and tested across 50+ user interactions.
2025
→

Everything else

Retrieval experiments, agent prototypes and coursework.
GitHub
03 / Experience
01

AI Engineer at Calyb AI

Backend intelligence layer for enterprise automation. Graph ontology over workflows, steps and API references. A supervisor coordinating quantitative and qualitative agents. PageIndex retrieval in place of plain vector search.
Jan — Aug 2026
02

AI/ML Intern at Divtechnosoft

Context aware response generation with LangChain and RAG. Integrated the Twilio WhatsApp Business API for real time chat. Ahmedabad.
May — Jul 2025
03

B.Tech in ICT, Dhirubhai Ambani University

Gandhinagar, India. CPI 7.64 out of 10.
Education
04 / Stack

Languages

  • Python
  • C++

AI / ML

  • LLM applications
  • Multi-agent systems
  • RAG
  • Prompt engineering

Systems

  • Knowledge graphs (Neo4j)
  • Ontology design
  • Workflow automation
  • PageIndex-based retrieval

Infrastructure

  • Agent isolation
  • Real-time streaming (SSE)
  • Vector databases
  • LangChain · LangGraph
05 / Contact
Let's talk

harsh.rathod0585@gmail.com

Savarkundla, Gujarat, India
I am looking for AI engineering roles, and I take contract work.

github.com/harshrathod0585
linkedin.com/in/harshrathod0585
+91 9510459823