slvDev
Featured
esp32-ai
A 28.9 million parameter language model runs on an $8 ESP32-S3 microcontroller entirely on-device, generating simple stories at about 9.5 tokens per second.
AI repository intelligence
FR-AI / ANALYSISWhy it is worth attention
It addresses a fundamental blind spot in current AI chat (only seeing final text) by leveraging decades of keystroke dynamics research (up to 75% accuracy in emotion classification) while enforcing a strict privacy-by-design policy: only timing metadata is stored, never the content.
Who it is for
- AI developers building empathetic conversational agents
- Researchers studying human-AI interaction and affective computing
- Product teams wanting to add emotional awareness to messaging apps
- Hobbyists experimenting with AI that can sense user hesitation
Use cases
- An AI companion that notices when you’ve typed and deleted something, then gently checks in
- Customer support AI that detects frustration from typing pauses before the user sends a message
- A writing assistant that offers encouragement when long pauses indicate writer’s block
- An AI that waits for the user to finish typing instead of interrupting mid-thought
Strengths
- Privacy-first design: frontend sends only a ping (no content), backend stores only timestamps – data structure cannot hold text
- Lightweight integration: adds ~10 lines of frontend code, one endpoint on the backend, and a small watcher script
- Low overhead: only introduces a few dozen tokens for real hesitations, ignores quick Q&A
- Backed by peer-reviewed research on keystroke dynamics and emotion recognition (75% accuracy across a dozen states)
Considerations
- Requires continuous pinging (every 4s) which might be perceived as tracking by users
- Emotion inference is probabilistic and based solely on rhythm; it may misread deliberate pauses or fast typos
- Watcher process and active-message AI capabilities may not be available or desirable in all applications
README quick start
快速开始（三步）
Description
Let your AI feel the hesitation in your typing. Rhythm only, never content.
Related repositories
Similar projects matched by category, topics, and programming language.
gavamedia
Featured
deltafin
Deltafin is a research project that runs the 2.8-trillion-parameter Mixture-of-Experts model Kimi K3 on a single Apple Silicon Mac (e.g., M1 Max with 64 GB) at about 16 seconds per token, using exact, reproducible inference with local or streaming expert loading.
jamesob
Featured
local-llm
A comprehensive guide for building and configuring a high-end local machine to run state-of-the-art LLMs, with detailed hardware choices, BIOS tuning, and Docker-based model serving.