Modelling
Regression, classification, NLP and transformer fine-tuning with careful evaluation.
Gourav Thakur
initialising
0%
Hi, I'm
AI & Machine Learning Engineer
I design and build end-to-end ML and AI applications — from data pipelines and model training to deep learning inference and FastAPI deployment.
Motivated MCA graduate with hands-on experience in Python, Machine Learning, and AI application development. Skilled in FastAPI, Scikit-Learn, TensorFlow, Pandas, and NumPy. Experienced in developing AI-powered applications, machine learning pipelines, and backend APIs. Passionate about solving real-world problems through AI and continuously learning emerging technologies.
AI / ML projects shipped
Tools & libraries used
Industry traineeships
MCA graduation year
Regression, classification, NLP and transformer fine-tuning with careful evaluation.
Cleaning, EDA and feature engineering pipelines built with Pandas and NumPy.
FastAPI services that serve models with typed schemas and predictable latency.
Modular, reproducible project architecture from ingestion to prediction pipeline.
Two industry traineeships spanning the full data-science-to-deep-learning arc.
AI/ML Trainee
Data Science Trainee
Grouped by how I actually use them — from language fundamentals to model evaluation.
Featured technologies
End-to-end machine learning and deep learning applications — data in, decisions out.

01 — Deep Learning Multi-Class Emotion Classifier
An end-to-end NLP system that detects emotion from raw text using deep learning. The pipeline loads a 20k+ sample Hugging Face dataset, cleans and tokenizes the text, handles class imbalance, and benchmarks RNN, LSTM, GRU and BiGRU architectures. The best BiGRU model reached 92.25% test accuracy. The trained model is served through a FastAPI inference API and consumed by a React frontend for real-time emotion prediction.
92.25%
Test Accuracy
20k+
Text Samples
4
Architectures
FastAPI + React
Deployment
Dataset & EDA
Loaded 20k+ text samples, explored class distribution and cleaned noise.
Text Preprocessing
Tokenized text, padded sequences and prepared embedding-ready inputs.
Model Benchmarking
Trained and compared RNN, LSTM, GRU and BiGRU architectures.
Imbalance Handling
Balanced classes during training to improve minority-emotion recall.
Evaluation
Compared accuracy, loss and confusion matrices to select the best model.
Serving
Deployed the BiGRU model behind a FastAPI endpoint for live inference.
Frontend
Built a React UI that calls the API and displays predicted emotion in real time.

02 — Machine Learning / Deep Learning
An AI-based skin disease classification project that uses deep learning and image processing techniques to classify different types of skin diseases from images. The project includes image preprocessing, exploratory analysis, model training, evaluation, and prediction.

03 — End-to-end regression pipeline
A production-shaped machine learning system that estimates the nightly price of an Airbnb listing from its location, capacity, amenities and host signals. The project is structured as independent, testable components — ingestion, transformation, training and inference — wired together by configurable pipelines and served through a FastAPI prediction endpoint, so retraining on fresh data never requires touching application code.
0.87
R² Score
↓ 24%
RMSE
5
Models compared
5
CV folds
Data Ingestion
Raw listings loaded, split and versioned into train/test artifacts.
Preprocessing
Null handling, outlier clipping, scaling and categorical encoding.
Feature Engineering
Geo-clustering, amenity counts, host-tenure and room-type signals.
Model Training
Linear, Random Forest and XGBoost regressors benchmarked side by side.
Tuning
GridSearchCV cross-validated search over the best-performing estimator.
Serving
Serialized pipeline loaded by FastAPI for real-time price predictions.
2024 – 2026
Sardar Patel University, Mandi
2020 – 2023
Rabindranath Tagore Government Degree College, Sarkaghat
Open to AI/ML engineering roles, internships and collaborations. I usually reply within a day.