01 · pycleanframe  ·  Open-Source Python Library for DataFrame Cleaning  

Published an open-source Python package on PyPI that simplifies DataFrame preprocessing using decorators.

Designed to eliminate repetitive data-cleaning code by providing reusable, configurable decorators for missing-value handling, normalization, validation, and preprocessing workflows.

from cleanframe import cleanframe

@cleanframe(drop_nulls=True, normalize=["age", "salary"])
def process(df):
    return df.groupby("department").mean()

• Published on PyPI
• Decorator-based API
• Reusable preprocessing pipelines
• Data validation & cleaning
• Lightweight and production-friendly

Python · Pandas · Decorators · Open Source · PyPI

02 · Dog Vision  ·  End-to-End Multi-Class Dog Breed Classification  

Built an end-to-end computer vision pipeline using TensorFlow and Transfer Learning to classify 120 dog breeds from images.

Trained on 10,000+ labeled images with an optimized tf.data pipeline, TensorBoard logging, EarlyStopping, GPU training, and Kaggle submission generation.

• MobileNetV2 Transfer Learning
• TensorFlow tf.data
• Image preprocessing & normalization
• TensorBoard
• EarlyStopping
• Kaggle submission pipeline

TensorFlow · Keras · Computer Vision · Transfer Learning · TensorBoard

03 · Resume Shortlisting System  ·  NLP-based Candidate Ranking  

Developed an end-to-end NLP system that parses resumes, extracts information, and ranks candidates against job descriptions using text similarity techniques.

• Resume parsing
• TF-IDF vectorization
• Candidate ranking
• NLP preprocessing
• Similarity matching

Python · NLP · Scikit-Learn · Pandas

04 · Blue Book for Bulldozers  ·  Kaggle Regression Project  

Built a complete regression pipeline for predicting heavy equipment auction prices using Kaggle's Blue Book for Bulldozers dataset.

• Feature engineering
• Regression modeling
• Cross-validation
• Model evaluation

Competition Metric: RMSLE = 0.24609

Python · Scikit-Learn · Regression · Pandas

Developed a complete machine learning workflow using the UCI Heart Disease dataset, including preprocessing, feature engineering, model comparison, and hyperparameter tuning.

• Logistic Regression
• K-Nearest Neighbors
• Random Forest

Python · Scikit-Learn · Pandas · NumPy

Built a predictive maintenance system that analyzes historical equipment data to forecast failures before breakdown, improving maintenance planning.

Python · Scikit-Learn · Pandas · Feature Engineering

07 · PentamedConsorsio  ·  Healthcare & Vaccination Tracking Platform  

Developed a full-stack healthcare platform for vaccination record management during UHACK 3.0, where the project finished Top 5 out of 198 teams.

Python · JavaScript · HTML · CSS · MongoDB

08 · Shopping API  ·  Transactional E-Commerce Backend  

A backend for a small shop: categories, products, orders, and stock that never goes negative — even when two people grab the last unit at the same moment.

• 6-table relational schema (users, categories, products, inventory, orders, order_items) with the reasoning written up in an ER doc
• Stock decrement done as one atomic SQL statement; the loser gets a 409 insufficient_stock and the whole order rolls back
• Concurrent-request test that proves exactly one of two simultaneous orders wins
• Alembic migrations for every schema change, including a real hotfix: a timezone bug that SQLite hid and real Postgres exposed
• Router → service → repository layering, Docker Compose with healthcheck-gated startup

Python · FastAPI · SQLAlchemy 2.0 · PostgreSQL · Alembic · Docker

🔗 View repo

09 · Secure API  ·  JWT Auth, Redis & Background Workers  

The production layer the first two projects skipped: authentication, authorization, rate limiting, caching, and work that happens off the request path.

• JWT login with bcrypt-hashed passwords, and identical errors for "no such user" vs "wrong password" so usernames can't be enumerated
• Admin / user roles: 401 when we don't know who you are, 403 when we do and the answer is no
• Redis rate limiter built on atomic INCR, so limits hold across multiple API instances (a Python dict wouldn't)
• Cache-aside endpoint with TTL, plus a Celery worker that sends the welcome email outside the HTTP request
• 4-service Docker Compose stack (API + Postgres + Redis + worker), tests on fakeredis, GitHub Actions CI

Python · FastAPI · JWT · Redis · Celery · PostgreSQL · GitHub Actions

🔗 View repo

10 · Ping / User API  ·  REST + MongoDB Fundamentals  

Where the request path finally clicked: validation, layered architecture, real database-enforced constraints, and an environment a fresh clone can start with one command.

• Users and Notes CRUD on MongoDB, with unique indexes on username and email returning 409 Conflict
• Pagination, tag filtering and whitelisted sorting, backed by a compound (tags, created_at) index
• Verified with explain(): index scan (IXSCAN) instead of a full collection scan at 5,000 documents
• Centralized error contract, structured logging, env-based config
• docker compose up --build from a fresh clone, no manual steps

Python · FastAPI · MongoDB · Pydantic · Docker

🔗 View repo