Skip to content
DSDhruv Sandhu
Machine Learning Engineer · IIT Bhubaneswar

Building GenAI systems from research to production.

I'm Dhruv Sandhu, a dual-degree student at IIT Bhubaneswar working on NLP, graph learning and efficient models. Previously ML at Jazzee Technologies and research at IIT Delhi — with hands-on MLOps and a knack for model compression.

Years in ML
3+

Years in ML

Projects built
30+

Projects built

Industry & research internships
3

Industry & research internships

ML club members led
100+

ML club members led

Dhruv Sandhu

NLP · Computer Vision · GNNs

ML Eng.

Fig. 00 — Gradient descent with momentum

PyTorchTransformersLangChainLangGraphGraph Neural NetworksKnowledge DistillationMLflowAirflowDockerAWS SageMakerLiveKitSupabasePyTorchTransformersLangChainLangGraphGraph Neural NetworksKnowledge DistillationMLflowAirflowDockerAWS SageMakerLiveKitSupabase

01About

Turning data into impactful AI products.

Dhruv Sandhu
Dhruv SandhuIIT Bhubaneswar

Hi, I'm Dhruv — a Machine Learning Engineer with a deep passion for using data to solve complex problems and build AI-driven applications that matter.

I thrive on challenges that blend theory with practical implementation. My work spans NLP, Computer Vision and Graph Neural Networks, and I'm always eager to explore new frontiers in AI.

Beyond code, you'll find me on the badminton court — it keeps my perspective fresh. I believe in continuous learning and collaboration to build solutions that work efficiently and contribute positively.

Education
B.Tech + M.Tech, IIT Bhubaneswar
Focus
NLP, Computer Vision, Graph Neural Networks
Previously
Jazzee Technologies · IIT Delhi · Coforge
Off the clock
Badminton

What I bring

Deep technical expertise

A track record of developing and deploying high-impact ML solutions across domains.

Inventive problem-solving

Creative approaches to complex challenges — turning hard problems into opportunities.

Collaborative by default

Clear communication and a strong belief in teamwork to build robust, scalable systems.

Business-oriented

Solutions that are technically sound and also drive tangible business value.

Clean, scalable code

Efficient, well-documented code that is easy to extend and stands the test of time.

Always learning

Staying at the forefront of AI — constantly learning and applying new techniques.

02Focus areas

Where my work connects.

Five areas I keep coming back to, and the threads between them. Each community in the graph is a focus area; the pulses trace messages passing along its edges.

Fig. 01 — Knowledge graph

03Skills

A toolkit for the full ML lifecycle.

From research prototypes to monitored production services — the languages, frameworks and infrastructure I work with day to day.

AI & Machine Learning

Training, fine-tuning and distilling models — from transformers to graph networks and LLM agents.

  • PyTorch
  • TensorFlow
  • Scikit-learn
  • LangChain
  • LangGraph

Languages

Python first, with systems and competitive programming roots.

  • Python
  • C++
  • Java

Data

Relational and document stores for products and pipelines.

  • PostgreSQL
  • SQL
  • MongoDB
  • Supabase
  • Firebase

MLOps & Cloud

Reproducible experiments, orchestrated pipelines and containerised deployments on AWS.

  • Docker
  • AWS
  • AWS SageMaker
  • MLflow
  • DVC
  • DagsHub
  • Apache Airflow
  • Astronomer
  • CI/CD

Tooling

The everyday workflow.

  • Git
  • GitHub
  • VS Code
  • Postman

04Selected work

Models, built from first principles.

All repositories on GitHub
01

Large Language Models

Llama 2, from scratch

A ground-up implementation of the Llama 2 architecture with KV caching, Rotary Positional Embeddings and Grouped-Query Attention — plus multiple decoding strategies including beam search and greedy sampling.

  • Python
  • PyTorch
  • LLM
  • Transformers
02

Model Compression

Knowledge Distillation

Distils a teacher model into a smaller student, followed by instruction fine-tuning. The student can be a layer-pruned or width-pruned teacher, or an entirely different architecture.

  • PyTorch
  • SFT
  • Pruning
03

Graph Learning · Paper Implementation

GraTeD-MLP

Implements GraTeD-MLP distillation: compressing a Graph Transformer into a vanilla MLP while retaining the benefits of its attention matrix over complex graph data.

  • PyTorch
  • GNN
  • Distillation

05Experience

Where I've shipped and researched.

  1. Jun 2025 — Jul 2025

    Jazzee Technologies

    ML Engineer

    • Engineered an AI-powered interview platform with real-time video and audio via LiveKit.
    • Designed ML/NLP pipelines for automated evaluation, transcript analysis and dynamic reports.
    • Built secure, scalable AWS infrastructure with Docker; integrated Supabase for data.
    • Delivered admin dashboards and candidate portals, and implemented cheating detection to ensure integrity.
    • NLP
    • LiveKit
    • AWS
    • Docker
    • Supabase
  2. May 2024 — Jul 2024

    IIT Delhi

    Summer Research Intern

    • Analysed multi-site fMRI data (healthy controls vs. MDD) to study inter-region correlations.
    • Applied the AAC atlas in a federated setting with Graph Neural Networks.
    • Worked with GCN/GAT, mixture-of-experts (MoE) and adversarial domain adaptation.
    • GNN
    • Federated Learning
    • MoE
  3. May 2023 — Jul 2023

    Coforge Ltd, Noida

    NLP Research Intern

    • Built an end-to-end pipeline to extract feedback from PDFs and run sentiment analysis for retention and upsell insights.
    • Experimented with embeddings, ML models and optimisation using BERT and RoBERTa.
    • Explored named-entity recognition with BiLSTM-CRF.
    • Transformers
    • BERT
    • NER

Education

  • B.Tech + M.Tech (Dual Degree), Mechanical Engineering

    2021 — Present

    Indian Institute of Technology, Bhubaneswar

    GPA 7.59 · Foundation in algorithms and applied ML

  • CBSE

    Until 2021

    Amity International School, Ghaziabad

    Active in technical clubs and competitions

Leadership

  • Data Science Head

    2023 — 2024

    Data Science Club, IIT Bhubaneswar

    Led a community of 100+ ML enthusiasts through workshops and mentoring.

  • Mechanical ML Lead

    2024

    General Championship, IIT Bhubaneswar

    Guided a 15+ member team to a bronze finish for the branch.

06Contact

Let's build something intelligent.

Whether it's an ML role, a research collaboration or an interesting problem, I'd love to hear about it. I usually reply within a day.

Send a message

It lands straight in my inbox.