wtf is ccmt-seat-estimate?

This project is a data pipeline and prediction tool built to help engineering students in India understand their odds during CCMT, the centralized counseling process used for M.Tech admissions. Rather than guessing based on last year's cutoff scores alone, it builds a small statistical model from several years of historical data to estimate the chances of getting into a specific program during a specific counseling round. The work happens in three stages. First, raw data is collected from the official government counseling portals, including opening and closing rank records going back to 2023, institute-to-program mapping tables, and seat matrix data showing available and vacant seats. The README explains a specific trick used to get this data: the official site normally splits results across many pages, but changing a particular part of the page's web address makes it render everything on one page instead, which makes scraping practical. Second, that raw, messy HTML and text data is cleaned and turned into structured spreadsheet files with a script called build_datasets.py, which strips out irrelevant page formatting and produces consistent tables linking institutes, programs, and categories together. Third, the actual prediction happens in predictor.py. Given a student's GATE exam score, subject paper, and category, it filters out programs the student is not eligible for, then applies linear regression across multiple years of cutoff data to project this year's likely cutoff for a chosen round. It goes further by calculating how much cutoffs have historically varied and using that to produce an actual probability, not just a guess, that a given score will clear the projected cutoff. If a program currently shows zero vacant seats in a special round, its odds are reduced accordingly, since getting in would depend on someone else unexpectedly dropping out. Running it requires cloning the repository, installing the listed Python packages, manually placing downloaded source files in the right folders, running the dataset builder, and then running the predictor with command line flags for score, paper, category, and round. Results print to the terminal and also save as a spreadsheet.