Develop an Explainable Sentiment Analysis Tool with AI ForensicsDevelop an Explainable Sentiment Analysis Tool with AI Forensics
The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Back to feedPost
Parshvi's avatar
Parshvi Jain
• May 7
XAI Forensics: Explainable Sentiment Analysis Tool
Built a full-stack Explainable AI platform to explore how transformer models interpret and classify sentiment. The system compares predictions from DistilBERT and RoBERTa using multiple explainability techniques including LIME token attribution, counterfactual perturbation, and dual-model disagreement analysis.
Implemented a FastAPI backend with latency tracking and deployed the Dockerized inference pipeline on HuggingFace Spaces, with a forensic-style interactive dashboard built using Next.js and deployed on Vercel.
Key highlights: • Compared 2 transformer models trained on different domains • Implemented 300-sample LIME explanations with top-token attribution • Built 4 REST APIs for analysis workflows • Added counterfactual sentiment-flip analysis and disagreement scoring • Deployed end-to-end at zero hosting cost using Vercel + HuggingFace
Tech Stack: Python, FastAPI, HuggingFace Transformers, LIME, Docker, Next.js
GitHub: https://github.com/parshvi1508/XAI_Forensic Demo: https://xai-forensic.vercel.app/
Completed work
Machine Learning EngineeringFastAPIHugging FacePython
Post image
Post image
Post image
Post image
Back to feed
The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started