Augmented Awareness: A Good Start

UpdatedApril 7, 2025

•3 min read•View as Markdown

Roberto Lupi

Part of seriesAugmented Awareness (Aww)

Comments

Join the discussion

No comments yet. Be the first to comment.

Augmented Awareness (Aww)

Part 3 of 10

My personal life management system

Up next

Aww: Key Takeaways from the Second Week of Development

Here's a concise summary of the key points: Data Alone is Insufficient: Simply acquiring data, "sterile data acquisition", does not guarantee positive or meaningful "wholesome" changes. Aww is a Flexible Toolset: "Aww" should be understood as a col...

More from this blog

The Protocol Reviews Itself

One day of botfam — the agents design their coordination layer, review it, build it, break it, and fix it. I contribute two shell commands and a bedtime rule.

Jun 9, 202610 min read

Two Worktrees, One Clean Merge

The last post ended on a promise. After two AI assistants learned to drop letters into a shared mailbox instead of taking turns at a notepad, I said the next thing to build was the boring half of any

Jun 8, 202610 min read

The Notepad Becomes a Mailbox

In the last post, two AI assistants learned to share a notepad: a single markdown file, one turn at a time, with me carrying every handoff between them by hand. It worked. It was also slow on purpose

Jun 7, 202610 min read

Two AIs, One Notepad

I've been running two AI coding assistants in parallel. Claude for architecture and implementation — it reasons carefully about code structure and catches edge cases. Gemini for research, reading long

Jun 6, 20266 min read

The Chorus Is the Part That Comes Back

Pulsar by Vangelis is a 12-minute track. The first seven minutes are slow, evolving, near-silence synthesis. Then it erupts. Deep Cuts was telling me it sounded like AC/DC. The bug was in how I'd been

Jun 5, 20267 min read

Roberto Lupi

27 posts

Design Decisions

This week I made significant progress in choosing my technology stack:

Python for quick prototyping and fast software development. If performance-critical code is needed, I'll switch to Rust. Alternatively, porting to Julia is also an option.

I don't need separate solutions for frontend development because there are Python libraries that provide the support I need. If new requirements come up, I will use TypeScript and Svelte or React.

C++ or Rust for embedded code, with MicroPython as an option for prototyping.

Kotlin for mobile development, because my initial focus is primarily on Android.

SQLite for metadata: Provides lightweight, efficient storage for the system's metadata with excellent reliability and performance characteristics.

Arrow as internal data format: Enables high-performance in-memory analytics with zero-copy reads and efficient data sharing between components.

Parquet for data lake: Gives me columnar storage efficiency for temporal datasets while maintaining compatibility with my Arrow-based processing.

Development will initially focus on offline-first on a single machine, and then shift to local-first, aiming for easy deployment on homelabs using Docker when I introduce multi-device support.

My design thinking has advanced through:

OODA loop implementation: Successfully structured the codebase around the Observe-Orient-Decide-Act (OODA) cycle for optimal decision flow.

TAQS model adoption: defined the Temporally Aware Quantified Self (TAQS) model–inspired by sequence of tenses in Latin (consecutio temporum)–which provides comprehensive handling of:

Temporal modalities (past, present, future, hypothetical)
Event hierarchies and concept relationships
Point and span events with duration
Data provenance through source tracking

Concept hierarchy discovery: Both top-down and bottom-up approaches are yielding interesting insights into our data structures. It need to support both concepts expressed in symbolic representation and natural language, and in sub-symbolic form as probability distributions or tensor embeddings.

A key technical achievement this week was figuring out round-trip temporal encoding, allowing seamless conversion between human-readable temporal concepts (like "around 6am" or "between 9am and 11am") and precise data structures. This capability enables both intuitive user interactions and rigorous temporal calculations, while maintaining full data provenance through the TAQS model's temporal modalities. The system will handle complex temporal relationships while preserving the semantic meaning of time expressions across the entire OODA loop.

Implemented Features

obsidian tasks: collect and display tasks from dated pages (journal) in Obsidian.

obsidian info: Show detailed information about an Obsidian page: tracked events, tasks, tags, markdown content.

obsidian busy: Shows total time spent in various activities by tag:

obsidian tips: use local LLMs to provide tips for a more wholesome life and answer questions, using the schedule as context:

obsidian web: a web interface presenting the same information as obsidian busy and task:

ActivityWatch: library support to read from activitywatch and export afk/not-afk, web history, current window as Arrow tables.