Instagram Review Analysis using an AI Agent
Fine-tuned Roman Urdu sentiment analysis on any public Instagram profile's comments, from username to insights dashboard.
SWE (AI/MLOps)NLPSentiment AnalysisBERTGradioWeb Scraping
Given only a public Instagram username, this agent scrapes the account's latest 30 posts and their comments (via Apify), classifies each comment's sentiment with a fine-tuned Roman Urdu BERT model (IndicBERT, re-labeled from a fine-grained scale to positive/negative/neutral via an SVM head), and surfaces the results, sentiment breakdown, trends, top keywords, through an interactive Gradio dashboard deployed on HuggingFace Spaces. Model weights are hosted in a private HuggingFace repo and pulled at runtime rather than committed to the codebase.
What I built
- Scraped the latest 30 posts and comments from any public Instagram profile using Apify's Instagram scraper, organized into a clean pandas pipeline.
- Fine-tuned IndicBERT (Roman Urdu) on two datasets, then re-labeled its output sentiment scale down to positive, negative, and neutral via an SVM classifier from scikit-learn.
- Built an interactive Gradio app, deployed as a HuggingFace Space, that takes just a username and returns a full sentiment breakdown, trend charts, and audience perception summary.
- Hosted fine-tuned model weights in a private HuggingFace model repo, pulled at runtime via an access token rather than committed to the repository.
- Built EDA visualizations (Matplotlib/Seaborn) surfacing sentiment trends, frequency distributions, and perception patterns per profile.
Tech stack
SWE (AI/MLOps)NLPSentiment AnalysisBERTGradioWeb Scraping