Aerle
An AI-native second brain
A place for your thoughts, with semantic search, voice notes, and AI agents that help connect the dots.

The problem
Getting thoughts into a second brain takes work: typing personal notes interrupts a brain dump, and sorting them into files and folders adds another task. Speech recognition tools such as Wispr make speaking instead of typing more practical, but transcription alone still leaves people to organize the notes manually.
The solution
Aerle combines voice transcription with LLM agents that organize notes and folders. Automatic note embedding and semantic retrieval let users ask questions about their own notes and review the sources attached to each answer.
Inspiration: How to build an AI-compatible second brain
My responsibility
I built the RAG pipeline, including automatic note embedding and semantic retrieval, and shipped 25+ API routes with content encryption, per-resource auth guards, and Google OAuth integration. The AI layer also supports transcription, enrichment, and tool-calling agents for notes and folders.
Constraints
- Personal notes are private content as well as context for AI responses. Access must be checked for each resource, not just at sign-in.
- Answers need to stay connected to the retrieved notes and expose their sources. The pipeline returns structured citations alongside the response.
Technical decisions
Capture and organize
A multi-provider AI layer combines Llama, GPT, and embeddings for voice transcription, content enrichment, and agents that organize notes and folders.
Find answers with sources
A RAG pipeline combines automatic note embedding with semantic retrieval. Responses use those notes as context and include structured source citations, so the answer and its sources can be reviewed together.
Protect user content
More than 25 API routes use per-resource authorization, AES-256-GCM encryption for user content, and Google OAuth integration.
Outcomes and evidence
- Voice transcription, note enrichment, and agents that organize notes and folders are implemented in the same application.
- The RAG pipeline embeds notes automatically and retrieves them semantically; 25+ API routes include the documented content-protection controls.
These are implemented capabilities, not measured user impact. Usage and answer-quality metrics are not published. This account is based on my resume and project notes; the live demo is linked below, and no public source repository is linked.