Open to ML / AI internships

I build machine learning systems from first principles to deployment.

Focused on Machine Learning, Deep Learning, NLP, LLMs, and building models from architecture to deployment.

View Projects Resume(CV)
Scale

2.3M

translation pairs

Transformer

34.82

BLEU score

Computer vision

99.7%

recall reported

Projects

5+

deployed ML systems

About Me


I'm Om Santosh Mishra, an inquisitive BSc.IT student at Smt. C.H.M. College focused on Machine Learning, Deep Learning, and NLP. I enjoy understanding how models work beneath the abstractions, not just using libraries and APIs.

I build by learning through implementation: from classical machine learning and computer vision to implementing Transformer architectures from scratch in PyTorch. My work involves training models, debugging failures, evaluating them properly, and turning working systems into usable applications.

  • Real-World Problem Solving

    Analyzing complex datasets to extract actionable insights and drive informed strategic decisions.

  • Data Science Engineering

    Python, SQL, Pandas, NumPy, Scikit-learn, and PyTorch.

  • Deep Learning & Transformers.

    Designing neural network architectures (CNNs, ANNs) and building Transformers from scratch and fine tuning them.

  • Interactive ML Applications

    Transforming predictive models into interactive, user-friendly web applications and dashboards using Streamlit and HuggingFace.

Skills in motion

The same stack from my existing portfolio, presented as a living system.

Python PyTorch TensorFlow Scikit-learn Pandas NumPy SQL Hugging Face Streamlit FastAPI Git GitHub Python PyTorch TensorFlow Scikit-learn Pandas NumPy SQL Hugging Face Streamlit FastAPI Git GitHub
Machine Learning Deep Learning NLP Transformers Computer Vision Feature Engineering Model Evaluation Deployment Machine Learning Deep Learning NLP Transformers Computer Vision Feature Engineering Model Evaluation Deployment
Machine learning

Scikit-learn · XGBoost / LightGBM · Feature Engineering & Evaluation · SHAP & Model Interpretability

Deep learning

PyTorch · CNNs & Neural Networks · Transfer Learning · Training & Optimization

NLP & Transformers

Tokenization & Sequence Modeling · Attention Mechanisms · Encoder-Decoder Transformers · NLP & LLM Fundamentals

Engineering

Python & SQL · Streamlit & FastAPI · Git & GitHub · Docker & Practical Deployment

Experience

FlyRank AI — Machine Learning Intern

Jul 2026 – Sep 2026

Completed a hands-on Machine Learning Internship focused on real-world search and content data, working through end-to-end workflows including data exploration, feature engineering, preprocessing, model development, and evaluation.

Developed a Search Intelligence capstone to identify pages at risk of declining search visibility using 90-day content and search signals. Built a ranking system with client-grouped validation, achieving 75% Precision@20 versus a 45% baseline, and converted predictions into ranked recommendations with interpretable reason codes and an actionable content playbook.

Projects

01Transformer Translation App
NLP · Transformers

English → French Transformer

Neural Machine Translation System

Encoder-Decoder architecture built from scratch for EN-to-FR translation. Achieved 34.82 BLEU score on 2.3M pairs with cross-attention heatmaps.

PyTorchNLPStreamlit
34.82BLEU
2.3Msentence pairs
from scratchPyTorch architecture
App GitHub
02PulmoScan AI App
Computer Vision · Deep Learning

PulmoScan AI

High-Sensitivity Pneumonia Screening

Diagnostic system fine-tuning DenseNet121 to detect pneumonia from X-rays, achieving 99.7% Recall. Features Grad-CAM explainability and a low-latency Streamlit UI.

PyTorchStreamlitOpenCV
99.7%reported recall
DenseNet121fine-tuned backbone
Grad-CAMexplainability
App GitHub
03Shriramartha Streamlit Interface
NLP

Ramartha Analysis

Semantic Analysis Application

End-to-end NLP system for semantic analysis of low-resource classical texts using Word2Vec and FastText. Features Unicode-native processing and real-time concept exploration.

NLPWord2VecHugging Face
Word2Vecembeddings
FastTextsemantic features
Unicodenative processing
App GitHub
04Urban Bike Demand App
Machine Learning

Urban Bike Demand

Demand Prediction System

Logistics optimization system achieving 93.5% R² with LightGBM. Uses cyclic encoding and weather analysis to proactively prevent station stockouts.

PythonSklearnStreamlit
93.5%reported R²
LightGBMprediction model
cyclictime encoding
Demo Code
05Telco Customer Churn App
Machine Learning

Telco Customer Churn

Customer Risk Prediction

End-to-end ML pipeline optimizing retention via Logistic Regression (81% Recall). Features a live Streamlit dashboard for real-time customer risk assessment.

PythonSklearnStreamlit
81%reported recall
Logisticregression baseline
liverisk assessment
Demo Code

Analytics

A minimal dashboard inspired by the reference: traffic, page views and a time-series view in the same dark visual language.

Visitors
0
↑ local
Page views
0
↑ local
Sessions
0
this browser
Pages / visit
0.0
live estimate
~/analytics.tsx
VisitorsPage Views

Local mode: counts are stored in this browser with localStorage. They do not represent global traffic unless connected to a server-side analytics provider.

Open source

GitHub

A live snapshot of the code behind the portfolio — repositories, activity, and the systems I keep building.

View profile

Contribution activity

Loading GitHub activity…

MonWedFri

Selected repositories

Loading…

Technical writing

Notes and engineering write-ups on AI, Data, and the systems I build.

Natural Language Processing

Building the Transformer Architecture from Scratch in PyTorch

A detailed engineering journey covering the implementation of the complete "Attention Is All You Need" architecture, large-scale training on 2.3M English-French sentence pairs, debugging numerical instability, beam search decoding, BLEU evaluation, and deployment with cross-attention visualizations.

Read Article
Deep Learning

LSTM Explained: From RNN Failure to Stable Sequence Learning

A deep dive into why vanilla RNNs fail and how LSTM uses gated memory to learn long-term dependencies.

Read Article
Guide

Machine Learning Made Simple

A beginner-friendly roadmap to understanding machine learning fundamentals.

Read Article
Computer Vision & CNNs

Building High-Sensitivity Pneumonia Screening

Deploying a fine-tuned DL model with Grad-CAM visualization and Streamlit.

Read Article
Self Learning

The Data Analysis Journey

How I went from zero Python knowledge to real-world data analysis projects.

Read Article

Contact

Have an ML problem worth building?

I'm open to ML / AI internships, research-oriented work, and collaborations where I can contribute to real systems and keep learning from strong engineers.

Email me LinkedIn GitHub

What you do makes a difference, and you have to decide what kind of difference you want to make.

Jane Goodall

Usually best reached by email or LinkedIn.