Clean and transform data with Pandas

Clean and reconcile a messy startup-funding export

Take the Indian Startup Funding dataset — a genuinely dirty real export with amounts stored as strings with commas, four different date formats, city names spelled several ways (Bangalore/Bengaluru/Banglore), investor lists crammed into one column, and duplicate rows from repeated scrapes. Produce one clean, analysis-ready table plus a data-quality report, as a reproducible Jupyter notebook that runs on the raw file with no manual pre-editing. Every judgement call — what you dropped, what you imputed, what you merged — gets written down.

Indian Startup Funding dataset on Kaggle — ~3k funding rounds with inconsistent dates, city spellings, and amounts stored as text

1. Profile the raw file: nulls per column, dtype surprises, duplicate candidates · ~2h
2. Parse the mixed date formats and coerce the text amounts to numbers, logging every failure · ~3h
3. Normalize city/industry/investor text with a mapping plus fuzzy matching, and de-duplicate · ~2.5h
4. Emit the data-quality report and the cleaned table, and write up the judgement calls · ~2h

• Notebook runs top-to-bottom on the raw file with no manual pre-editing required
• Produces a documented data-quality report: % nulls per column, duplicates found and removed, date-parse failures
• At least one fuzzy-matching or normalization step for inconsistent categorical text (agent names, statuses)
• Final clean table is exported and a short markdown section explains every judgment call made

Evidence a recruiter can check

• A data-quality report listing every issue found with counts — nulls per column, rows de-duplicated, dates that failed to parse and what you did with them
• The city/investor normalization mapping as a reviewable file, so a stranger can check whether 'Banglore → Bengaluru' style merges were right
• A notebook that runs top-to-bottom on the untouched raw file, with before/after row and null counts printed inline
• A short judgement-calls section naming what you dropped versus imputed, and why each choice could be wrong

Turned a 3k-row scraped funding export into an analysis-ready table with a reproducible Pandas notebook — normalizing four date formats, text-encoded amounts and duplicated city spellings, with every drop and merge documented.