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.
它采用控制环而非大型模型;确定性、可审计且资金安全,解决了市场平台动态抽成而不对低价值工作过度征税的需求。
A fast, deterministic controller that sets a per-transaction commission (take rate) to hit an aggregate target, while protecting the counterparty on the unattractive transactions.
Marketplaces usually want an overall take rate, say 20%, without charging the same percentage on
every transaction. A flat rate is simple but crude: it over-taxes low-value or hard-to-serve jobs
and leaves value on the table on attractive ones. take-rate-controller decides each
transaction's commission inside configured bounds so that the running effective take rate
converges on the target, lowering or zeroing commission on low-value or low-demand transactions
to protect earnings, and taking a little more on high-demand ones.
It is a control and optimization problem, so it is built as a control loop, not a large model: each decision is arithmetic plus a bounded, explainable modulation, returned in microseconds, fully auditable, and safe for money.
Scope note: a generic marketplace take-rate engine (rides, delivery, services). Deterministic at inference, with no LLM anywhere in the money path. This is the reusable engine, not a clone of any one platform's product.
flowchart LR
Req[amount, tax, segment] --> V[validate]
V --> C[controller\ntarget vs current effective]
C --> M[signal modulation\ndemand / supply / opportunity]
M --> K[clamp to segment bounds]
K --> D[decision: commission %, amount, payout]
Done[completed txns] -. async .-> Agg[running aggregates] --> C
D --> L[decision log]
C -. timeout / failure .-> FB[safe fallback] --> D
commission + payout == amount_after_tax is checked; the result
is hard-clamped so a payout can never go negative or out of bounds.Deterministic PI control loop that sets per-transaction marketplace commission to converge on an aggregate take-rate target, with integer money, hard bounds, and a guaranteed fallback.
根据分类、Topic 和编程语言匹配的相似项目。
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.
Cindy is an open-source AI agent that runs locally on your machine, integrates multiple AI harnesses and models, and provides memory, skills, and automation to perform real work in your projects and apps.