AI / Machine Learning Engineer · Fresher · 0 yrs

Devansh Rathore

AI / Machine Learning Engineer (Entry-Level)

Professional summary

Final-year Data Science graduate who builds and ships end-to-end ML — from data pipelines and feature engineering to PyTorch model training and a deployed inference API. Strong in the math behind the models (linear algebra, probability, optimization) and fluent in Python, scikit-learn and modern NLP with LLMs. Turns coursework into working systems, demonstrated across 3 deployed projects, a top-5% Kaggle finish and a fine-tuned transformer serving live predictions.

3

ML projects deployed live

Top 5%

Kaggle leaderboard finish

0.91

ROC-AUC, churn model

Skills

Languages & math

PythonSQLNumPyLinear algebraProbability & statisticsOptimization

ML & deep learning

PyTorchTensorFlowscikit-learnHugging Face TransformersXGBoostPandas

NLP & LLMs

TransformersFine-tuning (LoRA)EmbeddingsRAGspaCyPrompt engineering

MLOps & cloud

DockerFastAPIMLflowGitAWS (S3, EC2)GCP (Vertex AI)Streamlit

Work experience

Machine Learning Intern · Nimbus Analytics

Jan 2025Jun 2025

Remote

  • Built a scikit-learn churn-prediction pipeline on 180k customer records, reaching 0.91 ROC-AUC and surfacing the top 5 churn drivers for the growth team.
  • Engineered a reusable Pandas feature-pipeline that cut data-prep time ~40% and standardized inputs across 3 models.
  • Packaged a PyTorch sentiment model behind a FastAPI + Docker service deployed on AWS EC2, serving predictions at p95 under 180ms.
  • Tracked 25+ experiments in MLflow, making model comparisons reproducible and shrinking retraining turnaround from a day to under an hour.

Featured projects

+18% ROUGE-L
LexBrief — legal-document summarizer

NLP app that summarizes long contracts using a fine-tuned transformer.

  • Fine-tuned a Hugging Face transformer with LoRA on 4k contract clauses, lifting ROUGE-L 18% over the zero-shot baseline.
  • Shipped a Streamlit front end and FastAPI inference endpoint on GCP, returning summaries for 30-page PDFs in under 6 seconds.
PyTorchLLMsNLPGCP
Top 5% Kaggle
RetinaScan — diabetic retinopathy CV model

Computer-vision classifier grading retinal scans across 5 severity levels.

  • Trained a TensorFlow CNN on 35k labeled fundus images, reaching 0.89 quadratic-weighted kappa — top 5% on the Kaggle leaderboard.
  • Added Grad-CAM explainability overlays so reviewers can see the regions driving each prediction.
TensorFlowComputer VisionKaggle

Education

B.Tech Data Science & Artificial Intelligence

2026

IIIT Naya Raipur, Raipur

CGPA 8.9/10 · Coursework: Machine Learning, Deep Learning, NLP, Linear Algebra, Probability, Big Data Systems

Certifications

DeepLearning.AI Deep Learning Specialization

Coursera · 2025

TensorFlow Developer Certificate

Google · 2025

Achievements

Top 5% (of 3,400+ teams) — Kaggle diabetic-retinopathy detection challenge.

Open-source: 4 merged PRs to a Hugging Face dataset-tooling library (8k+ GitHub stars).

See the work in 3D

Explore Devansh's interactive WebGL portfolio — projects, skills and a way to get in touch.