InsightFlow

Inspiration

we wanted to tackle the common problem businesses face: delayed responses, low personalization, and unstructured customer feedback. inspired by existing ai chatbots and customer analytics tools on github, we decided to combine predictive churn insights with feedback prioritization into a single, intelligent workflow. the goal was to transform reactive customer support into proactive, insight-driven service.

What it does

predicts which customers are at risk of leaving (churn prediction) analyzes customer feedback from multiple channels (email, chat, social media) clusters and prioritizes feedback to surface actionable insights generates automated, personalized responses or retention actions provides a responsive dashboard for customer service teams to manage workflow efficiently

How we built it

frontend: next.js 14, react, typescript, tailwind css for a fast, responsive interface backend & db: supabase (postgresql, auth, realtime) to store customer & feedback data securely ai: openai api + azure openai for natural language understanding, feedback summarization, and action generation workflow: lightgbm churn model + faiss clustering for feedback analysis, integrated via fastapi endpoints deployment: vercel-ready for quick demo access

Challenges we ran into

handling multi-source feedback in a unified format while maintaining meaningful context balancing speed and accuracy for churn prediction on limited sample data generating actionable retention suggestions that felt human and personalized ensuring the dashboard remained intuitive while surfacing complex AI insights

Accomplishments that we're proud of

built a hybrid AI system that combines churn prediction with feedback clustering and automated action fully responsive and secure dashboard for easy team adoption created a system capable of real-world business impact in under a week

What we learned

integrating multiple AI workflows (prediction + summarization + action recommendation) is feasible in a short hackathon timeline end-to-end pipelines require careful handling of data preprocessing, embeddings, and model interpretability presenting complex AI insights effectively depends on a clear, simple, and visual dashboard

What's next for InsightFlow

expand multi-language support for global businesses integrate with real CRMs like hubspot or salesforce implement real-time notifications and workflow triggers for customer success teams enhance personalization with deeper user behavior modeling and A/B testing of retention actions

Built With

• auth
• claude
• next.js
• openapi
• postgresql
• react
• runtime
• supabase
• tailwindcss
• typescript