Octavo

Get answers from people who solved it - Search through real human experience, not web pages

Inspiration

We've all been there - asking Google "How do I break into product management from design?" and getting generic articles that don't capture the nuanced, lived experience of people who actually made that transition. Traditional search engines index static web pages, but the most valuable knowledge lives in human memory - the decisions, failures, and breakthroughs that shaped real careers and lives. What if we could search through collective human experience the same way we search the web? What if an AI could remember and learn from thousands of people's journeys, then provide wisdom that's impossible to Google? That's the vision behind Octavo - a search engine powered by collective memory that taps into the lived experiences of real people to answer questions that traditional search simply can't handle.

What it does

Octavo is a human-centered search engine powered by shared Letta memory agents that taps into the lived experiences of many to answer complex, nuanced questions better than Google ever could. Instead of indexing static web pages, Octavo creates a collective intelligence hive from real human memories and experiences. Core Innovation: Each user has their own Letta agent that stores their background, decisions, and life experiences like a personal journal. These individual memories are then anonymized and aggregated into specialized "sub-agents" for different domains (cs-careers, founder-journeys, grief-support, etc.), creating a searchable collective intelligence network. Key Features: 🧠 Crowdsourced Memory Responses: Get answers based on patterns from real people who've actually lived through similar situations - not generic blog posts or static content 🤖 Memory Agent Orchestration: Intelligent octopus-like system that automatically selects and coordinates the most relevant memory agents for each query 🎯 Auto Agent Selection: The system automatically identifies which collective memory domains (career transitions, startup journeys, etc.) are most relevant to your specific question 🔄 Version Control for Memory: Navigate through time with versioned memory branches - see "what worked for CS majors in 2020 vs 2024" or explore how advice has evolved 📊 Memory Exploration: Dive deep into the reasoning, tool calls, and specific memory blocks that power each answer with full transparency Example Queries That Showcase Octavo's Power: "I'm a first-gen psych major. How do I break into software?" → Gets advice from 86+ real people who made similar transitions "How did solo founders get their first 10 users?" → Synthesizes actual founder experiences, not startup blog theory "What's the best way to study for system design interviews?" → Real preparation strategies from people who actually passed these interviews "Show me what worked for career changers in 2020 vs 2024" → Temporal analysis of how strategies evolved post-pandemic The Octavo Difference: Traditional search gives you information. Octavo gives you wisdom from lived experience. As we say in our demo: "This isn't scraped text. This is based on actual lived memory, versioned and reasoning in real time. You can't Google this."

How we built it

Octavo combines three cutting-edge AI platforms to create something entirely new: 🗂️ Letta - Persistent Memory Agents Each user has their own Letta agent that stores personal experiences, decisions, and outcomes Memory blocks are anonymized and aggregated into domain-specific collective agents Persistent memory enables learning from patterns across thousands of experiences over time Version control for memory allows temporal queries and advice evolution tracking 🔄 Orkes - Intelligent Orchestration Complex workflows orchestrate multiple memory agents for cross-domain queries Parallel processing of memory retrieval, pattern recognition, and synthesis Dynamic agent selection based on query context and user needs Real-time workflow visualization shows the AI reasoning process ⚡ Groq - Lightning-Fast Inference Ultra-fast synthesis of memory patterns into coherent, actionable advice Real-time confidence scoring and pattern detection across memory blocks Dynamic response formatting based on query complexity and user context Sub-second processing of complex collective memory queries 🧠 Claude 4 - Advanced Reasoning Engine Powers our agents’ most complex thinking and synthesis tasks Excels at understanding nuance, abstract reasoning, and reflective summarization Enables emotionally intelligent advice and contextual awareness Seamlessly integrates with Letta’s memory to generate deeply human responses Tech Stack: Frontend: Next.js 14, TypeScript, Tailwind CSS, Framer Motion Memory Layer: Letta agents with PostgreSQL persistence Orchestration: Orkes workflow engine for multi-agent coordination Inference: Groq for rapid memory synthesis and pattern recognition

Challenges we ran into

Technical Challenges: Agent Coordination: Orchestrating multiple Letta agents with Orkes while maintaining response speed and coherence Temporal Consistency: Building version control for memory that allows meaningful temporal comparisons Collective Intelligence: Aggregating individual memories into collective wisdom without losing nuance Design Challenges: Trust Building: Users need to trust advice from "anonymous collective memory" - solved with transparency into reasoning and confidence scoring Query Complexity: Handling ambiguous, nuanced questions that don't have clear search parameters Integration Challenges: Multi-Platform Orchestration: Getting Letta, Orkes, and Groq to work seamlessly together Real-Time Processing: Maintaining sub-second response times while processing complex memory queries Scalability: Designing architecture that works for both hackathon demo and potential real-world deployment

Accomplishments we're proud of

Technical Achievements: First-of-its-kind: Built the world's first search engine powered by persistent collective memory Multi-Agent Mastery: Successfully orchestrated complex workflows across three cutting-edge AI platforms Performance: Achieved sub-second response times for complex collective memory queries

What we learned

About AI Orchestration: The power of combining specialized AI platforms creates emergent capabilities none could achieve alone Orkes workflows enable complex AI reasoning that goes far beyond simple API calls Persistent memory fundamentally changes how AI can serve users over time About Collective Intelligence: Human experiences, when properly aggregated, contain patterns that are impossible to find elsewhere Privacy and collective learning aren't mutually exclusive with the right technical approach Real wisdom comes from lived experience, not just information About the Future of Search: Traditional search is fundamentally limited by indexing static content The next evolution of search will be about accessing human knowledge and experience AI memory and orchestration will enable entirely new categories of applications

What's next

Immediate Roadmap (Post-Hackathon): Visual Memory: Enable image and document analysis to enrich memory blocks Memory Contribution: Allow users to actively contribute their experiences to grow collective intelligence Enterprise Deployment: Pilot with consulting firms and professional services companies Technical Evolution: Advanced Privacy: Implement homomorphic encryption for even stronger privacy guarantees Cross-Platform Memory: Enable memory agents to learn across different applications and contexts Predictive Intelligence: Use collective patterns to predict outcomes and suggest preventive actions Specialized Domains: Create expert collective memory for medical, legal, and technical domains