Edgar

Inspiration

I wanted to go beyond the typical "chat assistant" and create a digital companion that felt like a real character: a digital butler living in my pocket. Inspired by the "Second Brain" concept, my goal was to build a tool that doesn't just answer questions, but actually remembers and proactively assists me. That’s how Edgar was born.

What it does

Edgar is a high-performance personal assistant and a persistent memory system:

• Multimodal Perception: Edgar can see (analyzes images), hear (real-time voice transcription), and speak (using a premium British butler voice).
• "Second Brain" Memory: Using a vector database, Edgar stores information and retrieves it contextually through RAG (Retrieval Augmented Generation).
• Proactive Reminders: He analyzes speech, extracts dates and tasks, and automatically schedules notifications.
• Smart Web Summarizer: I can paste any link and Edgar extracts the essentials, summarizes them, and saves them to my long-term memory.
• Infinite Voice Journaling: A specialized mode for long-form thoughts where Edgar listens continuously, extracts key points, and generates actionable tasks.

How I built it

• Serverpod: The core backend engine handling business logic, authentication, and persistence.
• Flutter: For a fluid mobile experience with a high-end "Cyber Slate" design system.
• PostgreSQL & pgvector: I implemented native vector search within the Serverpod ecosystem to enable semantic memory.
• Google Gemini: Powers the central intelligence, multimodal analysis (Vision), and agentic reasoning capabilities (Tool Calling).

Challenges I ran into

• pgvector Support in Serverpod: Since Serverpod doesn't have native vector types in YAML yet, I had to implement custom SQL functions and projections to handle cosine similarity and vector operations.
• Continuous STT: Managing audio streams and silence detection in an emulator environment while maintaining a reactive UI was a significant technical hurdle.
• Visual Consistency: Achieving a polished glassmorphism effect with neon accents required fine-tuning Flutter's styling and custom painters.

Accomplishments that I'm proud of

• RAG Flow: The way Edgar can recall a specific detail from days ago and use it to answer today feels like magic.
• Premium Aesthetics: The "Cyber Slate" design elevates the app from a simple "wrapper" to a luxury tech product.
• Proactive Extraction: Getting the system to schedule reminders solely through voice, without me having to interact with a calendar manually.

What I learned

• I learned how to extend Serverpod with low-level database operations (Raw SQL) to leverage new technologies like vector databases.
• I discovered the nuances of prompt engineering for agentic behavior.
• I deepened my knowledge of UX patterns for multimodal assistants.

What's next for Edgar

• Integration Ecosystem: Connecting Edgar to Google Calendar, Gmail, and Notion.
• Multilingual Support: Expanding refined voice capabilities beyond British English.
• On-device Processing: Exploring Small Language Models (SLMs) for even faster responses.

Built With

• android
• antigravity
• dart
• flutter
• gemini
• pgvector
• postgresql
• serverpod