LGLOVANSHU GARG
SAY HELLO
PORTFOLIO — 2026GURUGRAM, INDIA

LOVANSHU
GARG

AGENTIC AI ENGINEER — LLM & BACKEND

I build multi-agent systems and RAG pipelines — and the production backends that keep them fast, observable, and honest. Reasoning, retrieval, and deployment, wired end to end.

VIEW PROJECTS →GITHUB ↗
AI/ML ENGINEER @ BYTEBELL — OPEN TO COLLABORATIONS
FIG. 01 — ORCHESTRATION TRACE

A TASK ENTERS THE SYSTEM — WATCH IT ROUTE BETWEEN AGENTS. MOVE YOUR CURSOR TO INSPECT A NODE. A SMALL PICTURE OF HOW MY PIPELINES THINK.

ABOUT
WHO I AM / HOW I GOT HERE

What began as curiosity — tinkering with HTML, CSS and small scripts — grew into a career building systems that reason, retrieve, and act.

Today I work on agentic AI at ByteBell: multi-agent workflows with LangChain, LlamaIndex and CrewAI, RAG pipelines backed by vector and graph stores, and the asynchronous data machinery that feeds them at the scale of a million-plus documents.

I care about the unglamorous parts that make AI real: secure APIs, message queues, containers, CI/CD. A clever agent that can't ship is a demo — I build the ones that run in production.

I also love sharing what I learn — contributing to open source, mentoring fellow developers, and grinding algorithms for the joy of it.

LOCATIONGurugram, India
CURRENTLYAI/ML Engineer @ ByteBell
FOCUSAgentic systems · RAG · Backend
EDUCATIONB.E. Computer Engineering ’25
ALSO INTOOpen source · Mentoring
500+ LEETCODE · TOP 10% GLOBAL1400+ CONTEST RATINGKAGGLE CERTIFIED — ML · PYTHON · SQLHACKTOBERFEST — MERGED PRS
EXPERIENCE
DATASHEET / REVERSE-CHRONOLOGICAL
AUG 2025 — PRESENT
REMOTE
CURRENT

AI/ML Engineer — Agentic AI & Backend

BYTEBELL
  • Design multi-agent workflows with CrewAI, LangChain and LlamaIndex — reasoning, planning and vector-backed memory wired into autonomous pipelines.
  • Built a production RAG system on Pinecone + Neo4j (vector + graph retrieval) that cut model hallucinations by 40%.
  • Engineered async ingestion with RabbitMQ and Celery — 1M+ documents cleaned, annotated and structured for model readiness.
  • Ship secure FastAPI / Flask microservices with OAuth and JWT, serving low-latency model outputs to frontend apps.
  • Containerize and deploy with Docker and Kubernetes; run CI/CD on AWS for smooth model rollouts.
LANGCHAINCREWAILLAMAINDEXPINECONENEO4JRABBITMQFASTAPIK8SAWS
FEB 2025 — JUN 2025
NOIDA, INDIA

Software Trainee

CADENCE DESIGN SYSTEMS
  • Built an NLP failure-prediction model(scikit-learn, TF-IDF) that reads failure logs and automates ticket assignment.
  • Optimized cloud-monitoring dashboards with JavaScript performance work — metrics, visualization, web architecture.
  • Developed a C++ utility for robust large-file management across system components.
  • Maintained rigorous code quality and versioning with Git and Perforce in an agile team.
PYTHONSCIKIT-LEARNTF-IDFC++JAVASCRIPTPERFORCE
2021 — 2024

The B.E. years — hackathons, freelance AI tools, and open-source contributions. Where the curiosity compounded.

PROJECTS
SELECTED BUILDS / OPEN SOURCE
FIG. 02 — DUAL-STORE RETRIEVALDOCSCHUNKSPINECONEVECTORNEO4JGRAPHLLMANSWERHALLUCINATIONS −40%INGEST → EMBED → RETRIEVE → GROUND

Iamyourbot

An end-to-end RAG information-extraction system — generative AI paired with vector search and Hugging Face embeddings for fast indexing and accurate querying over structured and unstructured text.

PYTHONRAGHUGGING FACEVECTOR SEARCH
VIEW LIVEVIEW CODE
FIG. 03 — SEQUENCE FORECASTHILOLSTM WINDOWNOWT+NT−30MARKETCNN + LSTM FORECAST

WiseTrader

Deep-learning time-series prediction — CNN and LSTM models in PyTorch and TensorFlow for pattern recognition in financial data, with the full data lifecycle managed from collection to feature engineering.

PYTORCHTENSORFLOWCNNLSTM
VIEW CODE
FIG. 04 — QUIZ-GATED PUSHAGENTQUIZ.JSONWRITES QUIZDIFFQ: WHY THIS?HUMAN ANSWERSGRADED LIVE — NEVER AUTO-PASSEDPASSEDgit pushPRE-PUSH HOOKCONTENT-ADDRESSED · NO SERVER AT PUSH TIMEUNREVIEWED PUSH → BLOCKED

quiz-axi

A git push gate for AI-authored diffs — after an agent finishes a change, it writes a short quiz about the diff, grades your answers live in the browser, and a Husky pre-push hook refuses to push anything that hasn't passed.

NODE.JSCLIGIT HOOKSAGENTIC WORKFLOWS
VIEW CODE
FIG. 05 — PERSONA GRAPHRAGCHUNKS DECAY, NEVER DELETEDYOUEXTRACTT=0.1PERSONA GRAPHGENERATET=0.7MATCH2ND-DEGREE NETWORK ONLY · CONSENT-GATEDTHE CONVERSATION IS THE PRODUCT

CalebX

A Telegram bot that builds a living persona from ordinary conversation — a local LLM extracts intent and sentiment into graph-native PersonaChunks, and GraphRAG traverses your 2nd-degree network to surface people, groups, and places that actually fit. No forms.

TYPESCRIPTGRAPHRAGHELIXDBOLLAMA
VIEW CODE
STACK
TOOLS I REACH FOR / GROUPED BY LAYER
AGENTIC AI & LLMS
  • LangChain
  • LlamaIndex
  • CrewAI
  • RAG · Multi-Agent Systems
  • Hugging Face · Transformers
  • Prompt & Tool Design
ML FRAMEWORKS
  • PyTorch
  • TensorFlow
  • Scikit-learn
  • Pandas · NumPy
  • CNNs · LSTMs
  • NLP · TF-IDF
BACKEND & APIS
  • FastAPI
  • Flask
  • RESTful Microservices
  • OAuth · JWT
  • Celery · RabbitMQ
  • Python PROFICIENT
CLOUD & MLOPS
  • AWS
  • Azure
  • Docker
  • Kubernetes
  • CI/CD Pipelines
  • Git · Perforce
DATA & SEARCH
  • Pinecone VECTOR
  • Neo4j GRAPH
  • Redis CACHE
  • ElasticSearch SEARCH
  • MongoDB · MySQL
LANGUAGES
  • Python
  • C++
  • SQL
  • JavaScript
  • HTML · CSS
CREDENTIALS
DEGREE / PROOF OF WORK
EDUCATION
B.E. Computer Engineering
J.C. BOSE UNIVERSITY OF SCIENCE & TECHNOLOGY, YMCA — FARIDABAD
2021 — 2025 · CGPA 8.2
PROOF OF WORK
Algorithms & Open Source
500+ LEETCODE PROBLEMS · TOP 10% GLOBAL · 1400+ RATING
KAGGLE CERTIFIED — ML · PYTHON · SQL · HACKTOBERFEST CONTRIBUTOR
CONTACT
INBOX ALWAYS OPEN

LET'S BUILD
SOMETHING AGENTIC.

Whether you have a question, a project, or just want to say hi — I'll do my best to get back to you. Reach out about agentic systems, RAG pipelines, or potential collaborations.

SAY HELLO →
GITHUBLINKEDINTWITTER