docs: add 'Intel resources worth knowing' section to both READMEs

Curated, opinionated list of upstream Intel-side resources behind the
stack, grouped into:

- OpenVINO toolkit (docs, openvino runtime, openvino.genai, NNCF)
- OpenVINO Model Server (model_server repo + demos, optimum-intel)
- Pre-converted models (OpenVINO HuggingFace org)
- Hardware (Core Ultra Series 2 family, Arc Graphics)
- Cloud / remote (Intel Tiber AI Cloud, optional)
- Community (OpenVINO blog, Intel Developer YouTube)

Placed between 'Project structure' and 'Troubleshooting' in both
README.md (EN) and README.es.md (ES); TOCs updated symmetrically.

Author: cartesiosson

README.es.md

@@ -54,6 +54,7 @@ Rendimiento medido: **~18-20 tokens/s** generando con Qwen3-8B INT4 sobre la iGP
 - [OVMS como backend de Claude Code](#ovms-como-backend-de-claude-code)
 - [Memoria y rendimiento](#memoria-y-rendimiento)
 - [Estructura](#estructura)
+- [Recursos Intel a tener en cuenta](#recursos-intel-a-tener-en-cuenta)
 - [Troubleshooting](#troubleshooting)
 - [Licencia](#licencia)
 - [Autor](#autor)
@@ -505,6 +506,73 @@ Sweet spot: pedirle "explica este fichero", "genera un test para esta función",
 └── openwebui-data/              # estado persistente de Open WebUI
 ```
 
+## Recursos Intel a tener en cuenta
+
+Lista corta y opinada de recursos upstream del lado Intel detrás de este stack
+— útil si quieres profundizar, cambiar componentes o contribuir.
+
+### Toolkit OpenVINO (el runtime debajo de todo)
+
+- 📖 [Documentación de OpenVINO](https://docs.openvino.ai/) — docs oficiales,
+  la fuente autorizada para comportamiento de plugins, ops soportadas y
+  hints de dispositivo.
+- 🐙 [`openvinotoolkit/openvino`](https://github.com/openvinotoolkit/openvino)
+  — el runtime C++/Python. Echa un ojo al [código fuente del plugin GPU](https://github.com/openvinotoolkit/openvino/tree/master/src/plugins/intel_gpu)
+  cuando un compile GPU falle de forma rara (el error `is_static()` que
+  tuvimos nosotros sale de ahí).
+- 🐙 [`openvinotoolkit/openvino.genai`](https://github.com/openvinotoolkit/openvino.genai)
+  — la capa runtime específica para LLMs (continuous batching, KV cache,
+  chat templates). Lo que OVMS usa por debajo.
+- 🐙 [`openvinotoolkit/nncf`](https://github.com/openvinotoolkit/nncf) —
+  Neural Network Compression Framework. Lee la
+  [documentación de weight compression](https://docs.openvino.ai/2024/openvino-workflow/model-optimization-guide/weight-compression.html)
+  para entender qué hace realmente `--weight-format int4 --group-size 64`.
+
+### OpenVINO Model Server (OVMS)
+
+- 🐙 [`openvinotoolkit/model_server`](https://github.com/openvinotoolkit/model_server)
+  — el servidor que usamos. El [directorio de demos](https://github.com/openvinotoolkit/model_server/tree/main/demos)
+  tiene ejemplos canónicos de `graph.pbtxt` para cada task (generación de
+  texto, embeddings, rerank, generación de imagen, VLMs). Cuando dudes,
+  copia de ahí.
+- 🛠 [`optimum-intel`](https://github.com/huggingface/optimum-intel) — el
+  puente con HuggingFace que convierte `Qwen/Qwen3-8B` en un IR de OpenVINO.
+  Nuestro `scripts/export-models.sh` es esencialmente un wrapper de
+  `optimum-cli export openvino`.
+
+### Modelos pre-convertidos
+
+- 🤗 [Organización `OpenVINO` en HuggingFace](https://huggingface.co/OpenVINO)
+  — modelos IR pre-convertidos oficialmente (Qwen, Llama, Phi, Mistral,
+  embeddings, etc.). Si no quieres esperar a la conversión local a INT4,
+  bájate uno de ahí y sáltate `scripts/export-models.sh`.
+
+### Hardware
+
+- 💻 [Procesadores Intel Core Ultra (Series 2)](https://www.intel.com/content/www/us/en/products/details/processors/core-ultra.html)
+  — la familia. Lunar Lake (Series 2) es para la que está tuneado este
+  stack, pero el mismo compose funciona en Meteor Lake y Arrow Lake H/HX
+  con la misma lógica de colocación iGPU/CPU.
+- 💻 [Intel Arc Graphics](https://www.intel.com/content/www/us/en/products/details/discrete-gpus/arc.html)
+  — la línea integrada (Arc 140V aquí) y la discreta hablan el mismo plugin
+  de OpenVINO. Si tienes un Arc A770/B580 discreto, reutilizas este mismo
+  stack con mucho más margen para modelos grandes.
+
+### Cloud / remoto (opcional)
+
+- ☁️ [Intel Tiber AI Cloud](https://www.intel.com/content/www/us/en/developer/tools/devcloud/services.html)
+  — la cloud de desarrolladores de Intel (antes Intel Developer Cloud).
+  Útil si quieres probar este mismo stack OVMS en una instancia mayor (Xeon
+  + GPU) antes de comprar hardware.
+
+### Comunidad
+
+- 📰 [Blog de OpenVINO](https://blog.openvino.ai/) — release notes, números
+  de rendimiento y anuncios de soporte de modelos. Suscríbete si vives en
+  este ecosistema.
+- 🎥 [Intel Developer en YouTube](https://www.youtube.com/@IntelSoftware) —
+  charlas técnicas sobre OpenVINO/OVMS y conferencias.
+
 ## Troubleshooting
 
 - **`/dev/dri` no existe en WSL**: actualiza Windows + drivers Intel Arc. Reinicia WSL: `wsl --shutdown`.

README.md

@@ -55,6 +55,7 @@ Measured throughput: **~18-20 tokens/s** generating with Qwen3-8B INT4 on the Ar
 - [OVMS as a Claude Code backend](#ovms-as-a-claude-code-backend)
 - [Memory and performance](#memory-and-performance)
 - [Project structure](#project-structure)
+- [Intel resources worth knowing](#intel-resources-worth-knowing)
 - [Troubleshooting](#troubleshooting)
 - [License](#license)
 - [Author](#author)
@@ -509,6 +510,72 @@ Sweet spot: "explain this file", "write a test for this function", "what does th
 └── openwebui-data/              # persistent Open WebUI state (gitignored)
 ```
 
+## Intel resources worth knowing
+
+A short, opinionated list of upstream Intel-side resources behind this stack
+— useful if you want to go deeper, swap components, or contribute.
+
+### OpenVINO toolkit (the runtime under everything)
+
+- 📖 [OpenVINO documentation](https://docs.openvino.ai/) — official docs, the
+  authoritative source for plugin behavior, supported ops and device hints.
+- 🐙 [`openvinotoolkit/openvino`](https://github.com/openvinotoolkit/openvino)
+  — the core C++/Python runtime. Browse the [GPU plugin source](https://github.com/openvinotoolkit/openvino/tree/master/src/plugins/intel_gpu)
+  when GPU compile fails in non-obvious ways (the error we hit with
+  `is_static()` was traced from there).
+- 🐙 [`openvinotoolkit/openvino.genai`](https://github.com/openvinotoolkit/openvino.genai)
+  — the LLM-specific runtime layer (continuous batching, KV cache, chat
+  templates). What OVMS uses under the hood.
+- 🐙 [`openvinotoolkit/nncf`](https://github.com/openvinotoolkit/nncf) —
+  Neural Network Compression Framework. Read the
+  [weight compression docs](https://docs.openvino.ai/2024/openvino-workflow/model-optimization-guide/weight-compression.html)
+  to understand what `--weight-format int4 --group-size 64` actually does.
+
+### OpenVINO Model Server (OVMS)
+
+- 🐙 [`openvinotoolkit/model_server`](https://github.com/openvinotoolkit/model_server)
+  — the server we run. The [demos directory](https://github.com/openvinotoolkit/model_server/tree/main/demos)
+  has canonical `graph.pbtxt` examples for every task (text generation,
+  embeddings, rerank, image generation, VLMs). When in doubt, copy from
+  there.
+- 🛠 [`optimum-intel`](https://github.com/huggingface/optimum-intel) — the
+  HuggingFace bridge that turns `Qwen/Qwen3-8B` into an OpenVINO IR.
+  `scripts/export-models.sh` is essentially a wrapper around its
+  `optimum-cli export openvino`.
+
+### Pre-converted models
+
+- 🤗 [`OpenVINO` organization on HuggingFace](https://huggingface.co/OpenVINO)
+  — official pre-converted OpenVINO IR models (Qwen, Llama, Phi, Mistral,
+  embedding models, etc.). If you don't want to wait for the local INT4
+  conversion, grab one of these directly and skip `scripts/export-models.sh`.
+
+### Hardware
+
+- 💻 [Intel Core Ultra processors (Series 2)](https://www.intel.com/content/www/us/en/products/details/processors/core-ultra.html)
+  — the family. Lunar Lake (Series 2) is what this stack was tuned for, but
+  the same compose file works on Meteor Lake and Arrow Lake H/HX with the
+  same iGPU/CPU placement logic.
+- 💻 [Intel Arc Graphics](https://www.intel.com/content/www/us/en/products/details/discrete-gpus/arc.html)
+  — both the integrated (Arc 140V here) and discrete Arc lineups speak the
+  same OpenVINO plugin. If you have a discrete Arc A770/B580, you can
+  reuse this exact stack with much more headroom for big models.
+
+### Cloud / remote (optional)
+
+- ☁️ [Intel Tiber AI Cloud](https://www.intel.com/content/www/us/en/developer/tools/devcloud/services.html)
+  — Intel's developer cloud (formerly Intel Developer Cloud). Useful if you
+  want to try the same OVMS stack on a larger Xeon + GPU instance before
+  buying hardware.
+
+### Community
+
+- 📰 [OpenVINO blog](https://blog.openvino.ai/) — release notes, perf
+  numbers and model-support announcements. Subscribe if you live in this
+  ecosystem.
+- 🎥 [Intel Developer YouTube](https://www.youtube.com/@IntelSoftware) —
+  OpenVINO/OVMS deep-dives and conference talks.
+
 ## Troubleshooting
 
 - **`/dev/dri` doesn't exist in WSL**: update Windows + Intel Arc drivers. Restart WSL: `wsl --shutdown`.