This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend
What I Built
For my close friend rahul, opening Slack and email in the morning isn't just routine—it's an obstacle course of cognitive friction. rahul has ADHD, and when a 4-paragraph rambling email or a 15-person chaotic Slack thread lands in their inbox, Task Paralysis sets in. The brain gets overwhelmed trying to separate social fluff and emotional subtext from actual deliverables. The notification gets minimized, and the task gets forgotten.
Existing AI tools (like ChatGPT or Claude) often make this worse: they are conversational, chat-heavy, and require sending sensitive work emails to third-party cloud servers—a major privacy violation for work communications.
Focus Friend is a minimalist, local-first executive function companion built specifically to conquer this problem. It distills chaotic walls of text into three digestible data points in high-contrast Neo-Brutalist cards:
- 🔵 THE VIBE (Cyan Card): A one-sentence plain-English emotional temperature reading (e.g., "The boss is pleased with progress and only asks for a routine slide polish"). This immediately disarms rejection sensitivity and anxiety.
- 🔴 THE DEADLINE (Coral Card): Isolates the exact calendar date or delivery window from atmospheric fluff.
- 🟢 THE ACTION ITEM (Lime Card): A single, imperative-verb bullet point (e.g., "Review slide 7 metrics and reply with confirmation"). 1-click copy into Todoist or Slack so rahul can take action and close the app.
Demo
- Live Full-Stack App on Render: focus-friend.onrender.com
- GitHub Repository: github.com/ezabhishek1/Focus-Friend (Anyone can test it instantly in their browser with the embedded Zero-Cloud Heuristic Engine and Overwhelm Presets, or run it locally connected to Microsoft Phi-3 via Ollama).
Code
ezabhishek1
/
Focus-Friend
**Focus Friend** is a local executive function assistant that uses open-weight AI to distill "walls of text" (emails, Slack threads) into three digestible data points: **Tone, Timeline, and Task.** By running locally, It ensures total privacy for sensitive work communications and provides a zero-latency experience to prevent focus-drift
⚡ Focus Friend — Local-First Executive Function Assistant
Submission for Hacktoberfest 2026: "Build for a Friend" Challenge
Dedicated to Alex ("The Overwhelmed Professional") and everyone navigating ADHD, neurodivergence, and cognitive overload in modern work environments.
🎨 Aesthetic Neo-Brutalism UI
Focus Friend is built with an unapologetic Neo-Brutalism UI Design System:
-
High-Contrast Geometry: Stark black borders (
3px/4px), tactile hard geometric drop shadows (box-shadow: 5px 5px 0px #121212), and zero blur. -
Visual Clarity for ADHD: Bold color-coding eliminates reading fatigue
- 🔵 The Vibe (Cyan
#38BDF8): Emotional context & tone to disarm rejection sensitivity. - 🔴 The Deadline (Coral
#FF6B6B): Isolates the true calendar milestone from noise. - 🟢 The Action Item (Lime
#A3E635): Single imperative bullet point for instant task initiation.
- 🔵 The Vibe (Cyan
- Calm Micro-Interactions: Non-startling auditory feedback synthesized entirely in-browser with the Web Audio API (warm major chords, soft click, pop) and focus-friendly breathing progress…
Stack & Architecture Highlights:
-
FastAPI & Uvicorn (REST API v1): Scalable Python backend with versioned endpoints (
/api/v1/extract,/api/v1/health,/api/v1/models,/api/v1/presets), threadpool concurrency, and global exception handlers. -
Local Inference via Ollama & Microsoft Phi-3: Runs open-weight
phi3:mini(2.3GB) strictly overlocalhost:11434with zero cloud telemetry. - Dual-Layer Zero-Cloud Heuristic Engine: Deterministic regex and sentiment rules in both Python and client-side JavaScript, ensuring 100% functionality even when offline.
- Aesthetic Neo-Brutalism UI: High-contrast 4px borders, hard drop shadows, ADHD color coding, and calm Web Audio API sound synthesis (no audio files downloaded).
- Centralized Reactive State Store: Observable pub/sub architecture with selector-based subscriptions and dedicated action dispatchers.
How I Built It
Focus Friend was engineered around local-first open weights, deterministic safety fallbacks, and sensory-friendly UX:
▼ [ExtractionService (FastAPI)] ├─► Tier 1: Local Ollama Daemon (Phi-3-mini-4k-instruct) via localhost:11434 └─► Tier 2: Zero-Cloud Heuristic NLP Engine (Deterministic Regex & Sentiment) │ ▼ [Structured 3-Card Distillation: The Vibe · The Deadline · The Action Item]
Why Does Open Innovation Matter?
Open-weight models and open innovation matter profoundly for applications that interact with the physical world:
- Uncompromised Privacy for Living Spaces: Urban exploration starts right from where you live—your apartment block, dorm, or neighborhood street. Proprietary closed APIs demand that you pipe your daily geographic context, walking routines, and photos to third-party corporate servers. With open-weight models, 0 bytes of personal habit data leave your device.
- Offline Resilience on Trails & Alleys: Real-world walks often take you into city parks, subway underpasses, or trail segments with poor cellular coverage. Closed API apps immediately fail when connection drops; open models running locally continue functioning in airplane mode.
- No Metered Billing Anxiety: Overcoming burnout shouldn't come with token quotas or surprise API bills. Open models empower builders to deliver free, resilient community utilities that anyone can run forever.
My Agent Session
This project was pair-programmed with Google DeepMind's Antigravity agentic coding harness, using the DevRelay gateway and official Agent Skills to inspect guidelines, establish the Neo-Brutalist design system, and manage deployment workflows.
Prize Categories
- Overall Open-Source AI Challenge
-
Best Use of Render: Deployed on Render using a native Blueprint (
render.yaml) as a Node.js web service running the Next.js App Router and dynamic server-side AI routes. -
Best Use of Gemma: Utilizes Google's open-weight
gemma:2b/gemma2:2bvia Ollama for lightweight, low-footprint quest distillation. - Best Use of GitHub Copilot: Used throughout the component modularization and test suite implementation.
Team Submissions: Built by @ezabhishek1.
Top comments (0)