Krishanu Bakshi

I'm a data analyst and machine learning practitioner based in Kolkata, India, with an M.Sc. in Applied Statistics and a B.Sc. in Statistics underneath it. My work sits at the intersection of statistics, machine learning, and practical automation — I like problems where I have to figure out why something isn't working before I can fix it, not just tune parameters until a number looks better.

Right now I'm a Trainee Data Analyst at DigitalSherpa, where I manage software asset lifecycle tracking across enterprise environments and automate Power BI and Python-based reporting — that work cut manual tracking effort by 40% and meaningfully improved visibility into operations that used to rely on manual dashboards.

Before that, I was a Research Project Intern at Jadavpur University, working on image classification with a real class-imbalance problem. I extracted hybrid features using VGG16 (CNN), HOG, and LBP, applied SMOTE to correct the imbalance, and pushed MLP recall from 52% to 77% — then optimized an XGBoost model to 99% recall. That project taught me to be skeptical of headline accuracy numbers and to actually dig into what a model gets wrong and why.

Outside of work, I build end-to-end ML projects to keep sharpening this. One recent one: a box office revenue prediction model that started with a badly overfit Linear Regression (R² of –167.70) — I diagnosed the problem through EDA and feature engineering rather than switching algorithms, and got it to 99.63% R². I also built a movie recommendation system on Databricks using TF-IDF, cosine similarity, and NLP embeddings on TMDB data, and trained an LSTM model that hit 83.87% accuracy.

Technically, I work mainly in Python and SQL, with scikit-learn, TensorFlow, Keras, and PyTorch for modeling, NLTK/SpaCy for NLP, and Power BI/Excel/Matplotlib/Seaborn for reporting and visualization. My statistics background (regression, classification, clustering, hypothesis testing, probability) gives me a level of rigor around model evaluation that I think is often missing when people only come at ML from a pure engineering angle.

I'm most interested in roles that combine data analysis with applied ML — particularly anywhere I get to evaluate model behavior critically, not just build and ship. I'm open to opportunities in data science, ML engineering, or AI evaluation/research, and I'm comfortable picking up new domains quickly (I've gone from image classification to NLP to time-series/regression work in the space of a year).