Echo

Inspiration

We've all seen it - the person who sits quietly in the corner during group discussions, the friend who struggles to reach out when they're going through something hard, the individual who feels like they don't belong anywhere. 1 in 3 people report feeling chronically lonely, yet traditional social platforms reward the loudest voices, not the quietest ones.

We wanted to flip that script. What if technology could meet people where they are, understand them without judgment, and gently guide them toward connection? That question became our North Star for this hackathon.

What It Does

Echo helps people with quieter voices feel heard through four interconnected features:

Personalized Emotional Support Chatbot

Unlike generic chatbots, ours analyzes your uploaded photos to understand your personality, hobbies, and life experiences. When you're struggling, it doesn't just offer platitudes - it reminds you of that beautiful trip to Paris, suggests activities aligned with your interests, and genuinely listens without judgment.

Photo-Based Personality Analysis

Users upload photos from their life, and our ML models extract insights about their hobbies, preferences, and personality traits. This creates a rich context that powers all other features.

Friend Matching

We connect users with similar people nearby based on genuine compatibility - shared hobbies, interests, and values - not superficial metrics. It's about finding your people.

Community Finder

For those who feel isolated due to language, religion, ethnicity, or other factors, we curate local events and communities pre-selected to match their specific needs.

How We Built It

Architecture:

Frontend: React with TypeScript for a responsive, accessible UI
Backend: Node.js/Express API with PostgreSQL database
Storage: Cloudflare R2 for secure photo storage
ML Pipeline:
- Custom image analysis model for personality/hobby extraction
- RAG-based chatbot using open-source models with Groq API fallback for quality assurance
- Sentiment/tone analyzer to ensure emotionally appropriate responses
APIs: Integration with Murf AI for text-to-speech accessibility features

We used a microservices approach where each feature operates independently but shares a common user context. The chatbot specifically leverages RAG (Retrieval-Augmented Generation) to pull relevant memories from analyzed photos, making conversations deeply personal.

Challenges We Faced

Balancing Privacy with Personalization

Analyzing personal photos is inherently sensitive. We spent significant time implementing privacy safeguards - photos are analyzed once, insights are extracted, and users control data retention.

Making the Chatbot Actually Helpful

Generic responses kill trust immediately. Our solution: implement a quality-checking layer where if the open-source model's response doesn't meet empathy thresholds, we fallback to Groq API. This two-tier system maintains quality without constant API costs.

Feature Cohesion

With four major features, we risked building four separate apps. We solved this by ensuring every feature feeds into a unified user profile - your analyzed photos inform friend matching, which connects to communities, which provides conversation topics for the chatbot.

The 24-Hour Reality

We coded through the night. At hour 20, our image analysis pipeline completely broke. Debugging ML models at 4 AM while running on Monster energy drinks taught us the importance of incremental testing (and proper nutrition).

What We Learned

Technical:

RAG architecture dramatically improves chatbot relevance when combined with user context
Multi-model fallback systems are worth the complexity for user-facing AI
TypeScript saves lives during sleep-deprived debugging sessions

Product:

Inclusion isn't about features - it's about intentional design for those who are typically overlooked
Personalization must be balanced with privacy from day one, not bolted on later
The best tech serves the quietest voices, not the loudest rooms