Skip to content
View jayzuccarelli's full-sized avatar
37 followers · 36 following

Achievements

Achievement: Pull Sharkx3Achievement: Pair Extraordinairex4Achievement: QuickdrawAchievement: YOLOAchievement: Arctic Code Vault Contributor

Achievements

Achievement: Pull Sharkx3Achievement: Pair Extraordinairex4Achievement: QuickdrawAchievement: YOLOAchievement: Arctic Code Vault Contributor

Block or report jayzuccarelli

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
jayzuccarelli/README.md

Jay Zuccarelli

building agentic systems and evals. shipping ml models in production.

🌐 jayzuccarelli.com · 𝕏 @jayzuccarelli · in jayzuccarelli

saccade · harness for proactive ambient agents. a cheap always-on model watches continuously and escalates to a larger one only on salience.

eden · claude agent tending a real hydroponic garden, with an esphome reflex tier holding safety deterministically underneath it.

memory-mcp · cross-llm memory over mcp. markdown files are the source of truth, so claude, chatgpt and cursor share one memory.

autofill · browser agent that fills any web form from a description of you, built on browser-use. you review and submit.

contributions

  • pytorch #193649 re-parented dynamo's dict_keys tracker so torch.compile matches eager
  • huggingface openenv #1074 stopped rate-limited sandbox installs from retrying and hiding the real error
  • huggingface trl #7450 fixed flops-per-token over-counting untied models: the embedding is a lookup, the lm head always one matmul
  • cuda (cccl) #11722 fixed cuda::discard_iterator returning a negated distance in nvidia's cuda c++ core libraries
  • vllm #58557 block-size errors now name each attention backend and the sizes it supports
  • opencv #30093 python typing stubs now import typing for Sequence return types
  • inspect_ai #4617 stopped a placeholder api key reaching the hugging face hub on model lookup
  • inspect_ai #5102 fixed chat templates that call dict methods on messages

Pinned Loading

  1. saccade saccade Public

    An open harness for proactive ambient agents — peripheral Glance, focused Focus.

    Python

  2. evernote-mcp evernote-mcp Public

    Self-hosted Evernote MCP server, native to the 2026-07-28 stateless spec

    Python

  3. autofill autofill Public

    Form-filling agent built on browser-use: describe yourself once, point it at any web form, and it fills every field automatically. You review and submit manually.

    Python 4 1

  4. memory-mcp memory-mcp Public

    Personal cross-LLM memory layer. Markdown files are the source of truth; the server exposes them over MCP so Claude/ChatGPT/Cursor/etc. share one memory.

    Python

  5. voice-pe-realtime voice-pe-realtime Public

    Self-hosted realtime speech-to-speech for the HA Voice PE: server-side OpenAI Realtime broker + dual-mode firmware, with Home Assistant control over MCP.

    Python

  6. ha-memory ha-memory Public

    Persistent memory for Home Assistant Assist, in plain markdown. No vector DB, no embeddings — works with Anthropic Claude, OpenAI, and any LLM conversation agent.

    Python 6