mybigday
- 27 skills
- 0 followers
- 6 hours ago last updated
- ▌ Device Test · mybigday bundleBuild and run the llama.rn example app on real hardware to validate native (cpp/, vendor/llama.cpp) changes end-to-end - iOS-on-Mac ("Designed for iPad" on Apple Silicon, real Metal GPU) and Android on a connected Snapdragon device with ggml-hexagon (HTP/NPU) enabled. Use this whenever the user wants to test on a real device or real GPU/NPU, says "run on my mac", "test on the phone/device", mentions hexagon, HTP, NPU, Metal GPU testing, or after C++ changes when unit tests pass but on-device verification is needed. Both flows compile the C++ from source, so cpp/ and vendor/ edits take effect.
- ▌ Upgrade React Native · mybigdayUpgrade React Native version in example/ directory using rn-diff-purge methodology. Use when the user wants to upgrade react-native, migrate to a new RN version, or update React Native dependencies in the example app. Handles diff retrieval, systematic file changes, dependency updates, and conflict resolution.
- ▌ Codec Dev · mybigdayGuidelines for developing the codec.cpp project, ensuring parity with llama.cpp architecture and conventions.
- ▌ Codec Op Dev · mybigdayGuide for adding or modifying ggml ops in codec.cpp with backend-safe patterns.
- ▌ Codec Model Dev · mybigdayGuide for adding a new codec model (GGUF conversion + ggml graph integration) in codec.cpp.
- ▌ Train Dpo · mybigdayDirect Preference Optimization (DPO) fine-tune with TRL `DPOTrainer`. Triggered when the user wants to align a model on preferences / pairwise comparisons / chosen-vs-rejected data, or improve an existing SFT checkpoint with a preference dataset.
- ▌ Train Mlx · mybigdayLoRA fine-tune a causal LM natively on Apple Silicon (M1/M2/M3/M4) via `mlx-lm`. Triggered when the user is on macOS arm64, mentions MLX, or wants better throughput than torch-MPS. Use this instead of `train-sft` when the host is a Mac with no NVIDIA GPU.
- ▌ Train Sft · mybigdaySupervised fine-tune a causal LM with TRL `SFTTrainer`. Triggered when the user wants to fine-tune / SFT / instruct-tune / chat-tune a model on conversational, prompt-completion, or text-formatted data. Enforces the literature-first → audit → smoke-test → scale workflow.
- ▌ Eval Model · mybigdayEvaluate a trained or downloaded language model with `lm-eval-harness` standard tasks (arc, hellaswag, gsm8k, mmlu, truthfulqa, ifeval, ...). Triggered when the user wants to benchmark, eval, or compare a model — pre- or post-training.
- ▌ Env Bootstrap · mybigdayRecreate the ml-intern-kit Python environment on a new machine (laptop, rented GPU box, Docker, fresh checkout). Triggered when the user is on a new host, sees ImportError on a core dep (torch/transformers/trl/peft/accelerate), or wants to install flash-attn / unsloth / bitsandbytes after the fact.
- ▌
- ▌ Inspect Dataset · mybigdayAudit a Hugging Face dataset before training to confirm splits, columns, format, sample rows, distributions, and duplicates. Triggered before any training/fine-tuning script runs, when a user mentions a new dataset, or when you hit a KeyError / format mismatch in a training job.
- ▌ Research Recipe · mybigdayRun a literature-first crawl before writing ANY ML training/fine-tuning/inference code. Spawns an Explore sub-agent that mines papers, citation graphs, methodology sections, and matched HF datasets to produce a ranked list of training recipes attributed to specific published results. Triggered when the user asks to fine-tune, train, or improve a model, or when the user names a task/benchmark and you need a recipe before coding.
- ▌ Rocm Strix Halo · mybigdaySet up training and inference on AMD Ryzen AI Max+ 395 / Strix Halo (gfx1151, RDNA 3.5) with TheRock nightly ROCm wheels. Triggered when the host has gfx1151, when `rocminfo` shows Strix Halo, or when the user mentions Strix Halo / Ryzen AI Max / gfx1151 / 128GB unified memory.
- ▌ AI Recipes · mybigday bundleUse when implementing a complete AI feature in a BRICKS app end-to-end, after generator and model choice. Proven wiring patterns with events, actions, and Property Bank state - voice kiosk (VAD → STT → LLM → TTS, or the assistant generator with MCP tools and file search), document Q&A / product finder (file ingestion → vectorStore → reranker → LLM), realtime transcription display, and AI image/video signage with the GenerativeMedia brick. Each recipe includes the generators and bricks to add, state shape, event wiring, offline behavior, and how to verify in the simulator and on device. Triggers on "build a voice assistant kiosk", "make a FAQ bot from these files", "live captions", "AI-generated visuals", "wire it up". Do NOT use for picking generators or models (use ai-generators / ai-model-selection first).
- ▌ PDF · mybigday bundleUse when the user brings a PDF into their BRICKS project and wants its content in the app — menus, price lists, schedules, catalogs, brochures, reports, or scanned documents. Extracts text on the design machine with a bundled script, routes the results into Data entries, Property Bank values, data-calculation sources, or media assets, and covers converting pages to images for slideshows. Also covers wiring runtime PDF ingestion (vectorStore RAG, parseDocument in data calculations) for apps that accept documents while running — note that mobile/TV launchers cannot parse PDF at runtime (web/desktop only). Triggers on "import this PDF", "read the menu from this PDF", "use this catalog", or any .pdf path in the prompt. Do NOT use for creating or exporting PDF files, or for Word docs, spreadsheets, and slide decks (use the docx / xlsx / pptx skills).
- ▌ AI Generators · mybigday bundleUse before adding any AI feature to a BRICKS app — chat, Q&A, voice interface, transcription, semantic search / RAG, or AI-generated media. Picks the right generator for the feature and target platform from the built-in catalog - local LLMs (llm/GGML, mlxLlm, neuropilotLlm, qnnLlm, onnxLlm, appleLlm), cloud LLMs (anthropicLlm, openaiLlm and compatible endpoints), speech-to-text (speechInference, realtimeTranscription, platformStt, appleStt, onnxStt), text-to-speech (ggmlTts, tts/ONNX, openaiTts, appleTts), vadInference, reranker, vectorStore, the assistant orchestrator, and the GenerativeMedia brick. Weighs offline requirements, privacy, latency, and per-device cost before wiring anything. Triggers on "add AI", "add a chatbot / voice assistant", "which generator should I use", "make it answer questions about…". Do NOT use for choosing the CTOR editor's own chat model (that is app settings, not the app being built).
- ▌ DOCX · mybigday bundleUse when the user brings a Word document (.docx) into their BRICKS project and wants its content in the app — menus, price lists, policies, product sheets, event programs, manuals, service guides. Extracts headings, paragraphs, lists, tables, and embedded images on the design machine with a bundled zero-dependency script; routes the results into Data entries, Property Bank values, data-calculation sources, or media assets, and can feed the vectorStore for document Q&A. Covers re-import when the document changes and runtime ingestion for apps that accept uploaded documents. Triggers on "import this Word document", "use this docx", "read the menu from this Word file", or a .docx path in the prompt. Do NOT use to create or edit Word files as the deliverable, and not for PDFs, spreadsheets, or slide decks (use the pdf / xlsx / pptx skills).
- ▌ PPTX · mybigday bundleUse when the user brings a presentation (.pptx) into their BRICKS project — company decks, menus designed as slides, event programs — and wants it on screen. Extracts per-slide text, titles, speaker notes, and embedded images on the design machine with a bundled zero-dependency script; then builds or updates the app — a Slideshow brick playlist, re-authored native layouts, or media-box assets. Covers slide-to-signage conversion (aspect ratio and safe areas, per-slide duration, loop behavior) and re-import when the deck changes. Triggers on "turn this deck into signage", "import these slides", "use this presentation", or a .pptx path in the prompt. Do NOT use to author new .pptx files as the deliverable, and not for PDFs, Word docs, or spreadsheets (use the pdf / docx / xlsx skills).
- ▌ XLSX · mybigday bundleUse when the user brings a spreadsheet (.xlsx, .csv, .tsv) into their BRICKS project and wants its data in the app — product lists, price tables, schedules, bookings, inventory, KPI sheets. Parses sheets into structured, typed rows on the design machine with a bundled zero-dependency script, shapes them, and imports as Data entries, Property Bank values, or data-calculation sources; large tables can also feed the vectorStore for semantic search. Covers re-import when the sheet changes and runtime ingestion for apps that accept uploaded sheets. Triggers on "import this Excel", "use this spreadsheet", "load the price list", or a spreadsheet path in the prompt. Do NOT use to create or edit spreadsheet files as the deliverable, and not for PDFs, Word docs, or slide decks (use the pdf / docx / pptx skills).
- ▌ AI Model Selection · mybigday bundleUse when choosing or changing the actual model for an AI feature in a BRICKS app — which GGUF, which quantization, which size — and when checking whether target devices can run it. Encodes hardware sizing rules - RAM budgets (the llm/GGML generator requires devices with more than 8GB RAM), acceleration per platform (Metal on M1+/A17+, OpenCL on Adreno 700+, Hexagon NPU on Snapdragon 8 Gen 1+, Vulkan/CUDA on desktop, WebGPU on web), context-length vs memory trade-offs, and quantization ladders. Uses the huggingface_search / huggingface_select tools to inspect GGUF metadata before committing, and lists starter models proven in BRICKS bundles (BricksDisplay/whisper-ggml for STT; Qwen 2.5 0.5B/1.5B, Gemma 2 2B class for constrained devices). Also covers when a cloud model is the better call. Triggers on "which model", "will this run on the device", "the model is too slow / too big", "pick a GGUF". Do NOT use for choosing generators (use ai-generators) or for the CTOR editor's own chat model.
- ▌ Template Creator · mybigdayUse when the user wants to extract a reusable template (prompt) from an existing BRICKS application. Analyzes the project's design, data flow, and key features, then produces a concise 5-10 item requirement list that can be used as a prompt to recreate a similar application. Triggers on "create template", "extract template", "generate template", "make a template from this app".
- ▌ Blender Pipeline · mybigday bundleUse when the user wants to create or modify 3D assets for a BRICKS app using Blender — product models, mascots, environments, 3D logos, exhibition pieces — or to convert existing models (FBX/OBJ/USD/.blend) into BRICKS-ready glTF. Drives the local Blender install headlessly (blender -b -P script.py) with bundled script templates for procedural scenes, format conversion, model inspection, optimization, and preview renders; optionally uses the Blender MCP server for live iterative modeling. Encodes BRICKS-safe export settings - .glb with embedded textures, NO Draco/KTX2/meshopt compression (the Scene3D loader does not support them), applied transforms, sane scale, and per-device polygon/texture budgets. Triggers on "make a 3D model", "create it in Blender", "convert this FBX", "optimize this model for devices". Requires Blender installed on this machine. Do NOT use for composing the in-app scene, lighting, or camera (use scene3d-authoring) or for scene behavior (use scene3d-interactions).
- ▌ Scene3d Authoring · mybigday bundleUse when adding or editing a 3D scene in a BRICKS app with the Scene3D brick — product viewers, showcases, menus with 3D flair, ambient scenes. Covers uploading .glb/.gltf/.usdz assets to the project media box (md5-addressed, preloaded for offline), declaring objects with transforms and per-node material overrides, lights (directional/point/spot/hemisphere), camera and orbit/pan-zoom controls, HDR/EXR environment backgrounds, tone mapping, bloom/FXAA/SSAO post-effects, and playing glTF animation clips. Ends with verifying in the simulator and deploying to devices (tvOS support is experimental). Triggers on "add a 3D scene", "show this model in the app", "3D product viewer", "set up lighting / camera / background". For creating the 3D assets themselves use blender-pipeline; for behavior, input, and games use scene3d-interactions.
- ▌ Bricks Telemetry · mybigday bundleAnalyse what real BRICKS devices and their users actually did, from `bricks device metrics` (uptime heartbeat, memory, disk) and `bricks activity-log` (every runtime event, plus screenshot history when enabled). Use for fault diagnosis (freeze, restart, wrong screen, failed payment or print), operator and end-user behaviour analysis (which bricks get pressed, dwell per screen, funnels and drop-off, sessions, busy hours, attended vs unattended), screen monitoring and state reconstruction (what was on screen minute by minute, with or without screenshots), fleet and usage reporting, and custom questions answered from the raw event stream. Also covers the query-cost rules that keep a pull from becoming a multi-hundred-megabyte download. Triggers on device troubleshooting, "why did the device do X", crash or freeze reports, uptime or memory questions, user/usage/engagement analysis, tap or touch analysis, session replay, screenshot history, activity log queries, or any request for a device report.
- ▌ Scene3d Interactions · mybigday bundleUse when making a Scene3D brick interactive — spin/inspect product viewers, click-to-explore demos, kiosk attract loops, or simple 3D mini-games. Covers the per-frame script (a pure expression evaluated every frame with time, dt, objects, camera, variables, keys, pointer in scope, returning a mutation map; statement-form scripts run once at mount), eventScript for key and pointer handling, raycast picking wired to BRICKS events and actions, and the imperative scriptMember API (addObject, removeObject, updateObject, setCamera, lookAt, playAnimation, stopAnimation, setBackground, setControls, screenshot, getSceneState, raycast). Includes movement and kinematics patterns — there is no physics engine, so no collisions or rigid bodies — plus performance budgets and connecting scene events to app state in the Property Bank. Triggers on "make it interactive", "clickable 3D objects", "rotate on touch", "simple 3D game", "attract loop". For static scene setup use scene3d-authoring.
- ▌ Maestro · mybigday bundleRun Maestro UI tests on the example app in iOS/Android simulators. Use when testing UI flows, verifying test suite results, automating the example app, running maestro commands, creating test flows, or validating changes via the on-device test runner. Triggers on requests involving Maestro, UI testing, running example tests, or verifying fixes on-device.