Python Projects in Bangalore

A Python application fetches daily OHLCV data for 50 NSE-listed stocks using the yfinance library and computes technical indicators — RSI, MACD, Bollinger Bands, and moving average crossovers — using the TA-Lib wrapper. A rule-based alert engine identifies buy and sell signals based on configurable indicator combinations and sends daily email summaries via smtplib. A Streamlit dashboard displays candlestick charts with indicator overlays and a signal history table. Backtesting on 2 years of historical data shows the MACD crossover strategy achieving a 14.3% annualised return on the Nifty 50 universe.

A Selenium and BeautifulSoup scraper monitors product prices across Amazon India and Flipkart for a user-defined list of products. Price history is stored in a SQLite database. A Flask web application displays price trend charts using Plotly and sends email alerts when a product's price drops below a user-set threshold. The scraper handles pagination, dynamic JavaScript rendering, and CAPTCHA detection with a polite crawl delay. A scheduler using APScheduler runs the scraper every 6 hours. The system tracked 100 products over 30 days, detecting 23 price drop events.

A command-line tool built with Python parses PDF and DOCX resumes using PyMuPDF and python-docx, extracts structured fields (name, email, skills, education, experience) using regex and spaCy NER, and scores each resume against a job description using cosine similarity of TF-IDF vectors. The tool processes a folder of 500 resumes in under 30 seconds and outputs a ranked CSV with match scores and extracted skill gaps. Unit tests cover 12 parsing scenarios including multi-column layouts, tables, and Unicode characters. The tool is packaged as a pip-installable CLI with argparse.

A FastAPI application wraps a YOLOv8 model fine-tuned on the COCO dataset to provide a REST endpoint for real-time object detection. Clients POST an image and receive a JSON response with bounding boxes, class labels, and confidence scores. The API processes images at 45 ms per frame on a CPU and 8 ms on a GPU. A React frontend allows users to upload images or stream webcam video and view annotated results in real time. The project includes Docker containerisation, a Swagger UI for API testing, and a performance benchmark comparing YOLOv8n, YOLOv8s, and YOLOv8m variants.

A packet capture tool built with Scapy captures live network traffic on a specified interface and extracts features — source/destination IP, port, protocol, packet size, and inter-arrival time — into a Pandas DataFrame. A pre-trained Random Forest classifier (trained on the NSL-KDD dataset) classifies each flow as normal or one of four attack types: DoS, probe, R2L, or U2R. Detected intrusions are logged to a SQLite database and displayed on a Flask dashboard with a live traffic graph. The tool achieves 97.2% classification accuracy on the NSL-KDD test set.

An Apache Airflow DAG orchestrates a daily ETL pipeline that ingests CSV files from the data.gov.in open data portal, validates schema and data quality using Great Expectations, transforms the data with Pandas, and loads it into a PostgreSQL database. The pipeline processes datasets from three domains: agriculture, health, and transport. A Metabase dashboard connected to the PostgreSQL database provides self-service analytics for non-technical users. The project demonstrates production-grade data engineering practices including idempotent tasks, retry logic, and Slack alerting on pipeline failure.

A domain-specific chatbot for a fictional e-commerce store is built using the Rasa open-source framework. The NLU model is trained on 800 annotated examples covering 12 intents: order status, return policy, product inquiry, shipping estimate, and escalation to human agent. Custom actions written in Python query a mock order database via REST API. The dialogue management model handles multi-turn conversations with slot filling. The chatbot is integrated into a React web interface and evaluated on a 100-conversation test set, achieving 91% intent accuracy and 87% task completion rate.

A plagiarism detection system processes student assignment submissions in PDF and DOCX format, extracts text, and computes pairwise similarity using TF-IDF cosine similarity, Jaccard similarity on 3-grams, and a BERT-based semantic similarity score. Submissions above a configurable similarity threshold are flagged with highlighted matching passages. The system also queries a web search API to detect internet plagiarism. A Flask web application allows instructors to upload a batch of submissions and download a similarity report. Processing 50 submissions takes under 2 minutes on a standard laptop.