I'm Anmol Kansal, a Data Science and AI/ML undergraduate specialising in Natural Language Processing. I build machine learning systems from first principles, experiment with language models, and develop practical backend applications.
A mix of NLP implementations, deep-learning experiments, backend systems and applied machine learning — with an emphasis on understanding what is happening under the hood.
Built an end-to-end word-embedding pipeline from first principles: regex tokenization, frequency-cut vocabulary, Mikolov subsampling, an O(1) negative sampler from the unigram0.75 distribution, Skip-Gram, CBOW and GloVe with weighted least squares over sparse co-occurrence data.
Implemented teacher–student learning on CIFAR-10 using a fine-tuned ResNet18 teacher and lightweight custom CNN students. Used KL-divergence with temperature scaling and tested architectural changes including global average pooling.
Full-stack project-management platform for creative marketing agencies. Built REST APIs and backend services covering projects, milestones, tasks, briefs, deliverables, budgets and timesheets, with RBAC, audit logging and dual-approval budget workflows.
End-to-end machine-learning pipeline predicting customer purchase value from multi-session clickstream data. Covered exploratory analysis, feature engineering, model selection and tuning as part of a Kaggle competition.
Built a web platform for brand–influencer collaboration with campaign creation, sponsorship requests, communication and progress tracking. Added Redis caching and Celery background jobs for notifications and periodic metric updates.
Analysed real-world sales and inventory data from a fashion business. Applied ABC/Pareto analysis, demand forecasting, sales-to-stock ratios and monthly inventory movement to identify purchasing and inventory optimisation opportunities.
Current work in NLP research, backed by hands-on backend engineering experience from an e-commerce intelligence project.
NLP research on LLM-based social agents. Reviewing literature around social-intelligence simulation and theory-of-mind evaluation of language models as groundwork for experiments.
Developed a real-time competitor strategy tracker for e-commerce using Playwright and Crawl4AI extraction pipelines, a FastAPI backend, MongoDB Atlas and Google Gemini API for analysing unstructured product data.
A practical stack spanning machine learning, NLP, backend development, data work and software engineering.
Python · SQL · Java · Statistics · Data Structures & Algorithms
PyTorch · Scikit-learn · TensorFlow · Pandas · NumPy · Deep Learning
NLP · Transformers · Tokenization · Text Processing · Prompt Engineering · Google Gemini API · Generative AI
FastAPI · Flask · SQLAlchemy · PostgreSQL · MongoDB · REST APIs
Git · GitHub · Jupyter · VS Code · API Testing · Redis · Celery
English · Hindi
Computer Science and AI/ML education with a growing focus on data science, NLP and deep learning.