Skip to content
View bernardogv's full-sized avatar

Block or report bernardogv

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
bernardogv/README.md

Bernardo Gallegos Vallejo

MTS Product Application Engineer at AMD, validating AI infrastructure. On my own time: upstream fixes to GPU stacks, Rust and inference tooling, and a GPU homelab run by agents.

bernardogv.com · LinkedIn · X

Upstream

23 pull requests to 10 open-source repos, 10 merged.

Project Merged Open Latest
llamastash/llamastash 4 0 fix(unit5): the vLLM UMA guard launches beside a tenant and reserves for the engine's own overhead
lemonade-sdk/lemonade 2 1 docs: extra_models_dir under /home needs a drop-in on deb/rpm
mostlygeek/llama-swap 2 0 config: accept video as an input and output modality
ROCm/ROCmValidationSuite 1 4 ci: check out the PR head commit so fork PRs build
ROCm/rocm-systems 1 0 fix(rocjitsu): avoid a GCC 15 free-nonheap-object false positive in build_cfg
bentoml/llm-optimizer 0 3 fix: stop waiting when the server exits; add --server-timeout
ROCm/TheRock 0 2 docs: add Ubuntu 26.04 notes to the environment setup guide
FoggedLens/deflock-app 0 1 fix(tiles): pause/cap tile retries when offline (#136)
ROCm/libhipcxx 0 1 fix(ci): derive LLVM_PATH from ROCM_PATH instead of hardcoding /opt/rocm
rust-lang/stdarch 0 1 amdgpu: document that ds_permute/ds_bpermute take a byte address

Updated daily · every PR →

Built with Claude Code

  • comfyui-mcp: 16-tool MCP server that drives a remote ComfyUI: graph-rewriting overrides, batching, curation, Python graph authoring.
  • proton-mcp-server: email for AI agents over Proton Bridge, with preview-before-send guardrails.
  • claude-homelab-plugin: a skill, slash commands, an operator subagent and an SSH MCP server with guardrails, gated by claude plugin eval suites on every release.
  • Local-model delegation: Claude scopes and verifies; a coder model on a DGX Spark writes the code in an isolated worktree; tests that must fail first decide what gets committed.

The lab

2× DGX Spark (GB10) · Strix Halo (Ryzen AI Max+ 395) · 200 Gb/s RoCE between the Sparks · a MAAS bare-metal fleet provisioned from Terraform PRs · 12U rack · every number in a dated casebook. Write-ups: spark-serving-bench.

Personal account. Views and projects are my own, not my employer's.

Pinned Loading

  1. proton-mcp-server proton-mcp-server Public

    MCP server for ProtonMail via Proton Bridge — gives AI agents full email management over IMAP/SMTP

    TypeScript 2

  2. deflock-vision deflock-vision Public

    Labeled ALPR camera dataset + zero-training vendor identification, toward deflock#120

    Python

  3. spark-serving-bench spark-serving-bench Public

    LLM serving on NVIDIA DGX Spark (GB10/sm_121a): benchmark harness + findings — SGLang serving recipe for sm_121a, prefix-caching gains, llama.cpp concurrency traps

    Python

  4. lemonade-sdk/lemonade lemonade-sdk/lemonade Public

    Lemonade helps users discover and run local AI apps by serving optimized LLMs right from their own GPUs and NPUs. Join our discord: https://discord.gg/5xXzkMu8Zk

    C++ 6k 539

  5. comfyui-mcp comfyui-mcp Public

    MCP server for driving a remote ComfyUI instance from an AI agent — 16 tools: queueing with graph-rewriting overrides, batching, model downloads, curation, and Python graph authoring via ComfyScript

    TypeScript 1

  6. claude-homelab-plugin claude-homelab-plugin Public

    Claude Code plugin for operating a GPU homelab: guarded SSH MCP server, fleet skill, operator subagent, audit hook, and eval suites

    JavaScript