
learn-agent
A collection of notes on building coding agents, derived from the production development of the desktop agent Reina, with each mechanism simplified into a zero-dependency, single-file Node.js demo.
它解决了2024年孟加拉国网络中断期间的实际问题,利用现有短信基础设施,无需双方预先安装应用,具有实测压缩性能和优雅降级能力。
An offline Bangla SMS app for the network a blackout leaves behind. It does three things, all on your phone, with no server and no internet:
Try it live · Slide deck · Deck as PDF — July Hackathon 2026, Track A (Crisis Tech).
New here? Watch the five-minute demo:
bornomala-final-5min.mp4. It walks through the app screen by screen — what an SMS segment actually is, why Bangla only gets 70 characters, how compression can cost less than one bit per letter, and what Emergency, Relay and the QR code are each for.
| Bornomala | raw UCS-2 (phones today) | gzip -9 | |
|---|---|---|---|
| Bits per Bangla character | 3.106 | 16.000 | 21.002 |
| Bangla characters per SMS segment | 361 | 70 | 53 |
| Of 5,000 held-out messages, fit in one segment | 100.0% | 98.2% | 99.5% |
Measured, not claimed — npm run bench, output in
benchmark.json.
In July 2024 the state cut 3G and 4G in Bangladesh. Broadband went dark. What survived was 2G — voice and SMS. People fell back to text.
But text messages were never designed for Bangla. SMS was built around an old character set that covers English letters and little else. Write a single Bangla letter and the whole message switches to a heavier format: 70 characters per message instead of English's 160. Speaking your own language costs more than twice as much to send, and fails more often when the towers are jammed — exactly when texting is all anyone has left.
Bangladesh has fought this fight before. 1952 was the right to speak Bangla. 2024 was the right to speak at all. This is the same fight one layer down, in the character encoding.
This is the honest question, and it
Offline Bangla SMS app — compress messages 5.15×, send plain-text emergency reports, relay many households in one SMS. No server, no internet.
根据分类、Topic 和编程语言匹配的相似项目。

A collection of notes on building coding agents, derived from the production development of the desktop agent Reina, with each mechanism simplified into a zero-dependency, single-file Node.js demo.
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.

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.