How systemhubtahir/data-cleaning-assistant works
Data Cleaning Assistant profiles a customer-call dataset, sends the profile—not raw rows—to an LLM for suggestions, and leaves decisions to a human before fixes are applied. Its stages are profiling, AI analysis, manual review, and cleaning. The reviewed suggestions are not automatically consumed by the apply script: the documented fixes are implemented directly in that script. The graph reflects documented workflow and sampled file operations; unsampled files are omitted.
Main entry points: clean_with_ai.py AI analysis · profile_data.py Profile data · apply_fixes.py Apply fixes · ai_suggestions.txt AI suggestions
Components 10
Dataset profiling
- Raw dataset
data/Customer_Call_List.xlsx - Profile data
profile_data.py
AI suggestions
- Profile report
reports/profile.json - AI analysis
clean_with_ai.py - AI suggestions
reports/ai_suggestions.txt
Manual cleaning
- Review suggestions
- Apply fixes
apply_fixes.py - Cleaned dataset
data/Cleaned_Customer_Call_List.csv
Other components
- Human operator
- Hugging Face API
Connections 12
- Human operator → Profile data: runs
README.md - Profile data → Raw dataset: reads
profile_data.py - Profile data → Profile report: writes
profile_data.py - Human operator → AI analysis: runs
README.md - AI analysis → Profile report: reads
clean_with_ai.py - AI analysis → Hugging Face API: requests suggestions
clean_with_ai.py - AI analysis → AI suggestions: writes
clean_with_ai.py - Human operator → Review suggestions: reviews
README.md - AI suggestions → Review suggestions: informs decisions
README.md - Human operator → Apply fixes: runs approved fixes
README.md - Apply fixes → Raw dataset: reads
apply_fixes.py - Apply fixes → Cleaned dataset: writes
apply_fixes.py