sites.json

{
  "SITES": {
    "kernelfusion": {
      "key": "kernelfusion",
      "url": "https://kernelfusion.dev",
      "domain": "kernelfusion.dev",
      "name": "kernelfusion.dev",
      "tagline": "Single-kernel fusion — sequential GPU dispatch chains collapsed into one dispatch.",
      "shortDesc": "The research line. Two published preprints and one npm SDK. The theory the applied projects build on.",
      "category": "Theory",
      "badge": "research",
      "githubRepo": "https://github.com/abgnydn/kernelfusion",
      "stats": [
        {
          "value": "DOI-archived",
          "label": "two preprints on Zenodo"
        },
        {
          "value": "@wgpu-fusion/core",
          "label": "npm SDK"
        },
        {
          "value": "WGSL",
          "label": "hand-written kernels"
        }
      ],
      "cta": "Read the research →"
    },
    "webgpudna": {
      "key": "webgpudna",
      "url": "https://webgpudna.com",
      "domain": "webgpudna.com",
      "name": "webgpudna.com",
      "tagline": "Geant4-DNA Monte Carlo in the browser. One fused dispatch per primary.",
      "shortDesc": "Electron track-structure simulation ported from the CNRS/IN2P3 Geant4-DNA toolkit to WebGPU. One thread per primary, full 10 keV history in a single for-loop. Radiolysis chemistry and DNA damage scoring live in a browser tab.",
      "category": "Radiobiology",
      "badge": "flagship",
      "githubRepo": "https://github.com/abgnydn/webgpu-dna",
      "stats": [
        {
          "value": "Geant4-DNA",
          "label": "physics ported"
        },
        {
          "value": "validated",
          "label": "radiolysis vs Karamitros 2011"
        },
        {
          "value": "browser-native",
          "label": "no install"
        }
      ],
      "cta": "See the simulation →"
    },
    "zerotvm": {
      "key": "zerotvm",
      "url": "https://zerotvm.com",
      "domain": "zerotvm.com",
      "name": "zerotvm.com",
      "tagline": "Phi-3-mini in the browser on hand-written WGSL kernels, no compiler.",
      "shortDesc": "Phi-3-mini running end-to-end in the browser on hand-written WGSL, replacing the TVM-autotuned shader set WebLLM compiles. 69.55 tok/s against WebLLM's 59.95 on identical weights in the same session, +16%.",
      "category": "LLM inference",
      "badge": "applied",
      "githubRepo": "https://github.com/abgnydn/zero-tvm",
      "stats": [
        {
          "value": "hand-written",
          "label": "every WGSL kernel"
        },
        {
          "value": "no compiler",
          "label": "no TVM, no codegen"
        },
        {
          "value": "69.55",
          "label": "tok/s (M2 Max, vs WebLLM same session)"
        }
      ],
      "cta": "Run it live →"
    },
    "gpubench": {
      "key": "gpubench",
      "url": "https://gpubench.dev",
      "domain": "gpubench.dev",
      "name": "gpubench.dev",
      "tagline": "How fast is your GPU in the browser?",
      "shortDesc": "Open WebGPU compute benchmarks — Rastrigin, N-body, Monte Carlo Pi, RL environments. Every submitted run is public, no cherry-picking.",
      "category": "Benchmarks",
      "badge": "companion",
      "githubRepo": "https://github.com/abgnydn/gpubench",
      "stats": [
        {
          "value": "no install",
          "label": "runs in the browser"
        },
        {
          "value": "public",
          "label": "every submitted run downloadable"
        }
      ],
      "cta": "Benchmark your GPU →"
    },
    "neuropulse": {
      "key": "neuropulse",
      "url": "https://neuropulse.live",
      "domain": "neuropulse.live",
      "name": "neuropulse.live",
      "tagline": "Watch Phi-3-mini think — every tensor rendered live from WebGPU.",
      "shortDesc": "A real forward pass of Phi-3-mini visualised tensor-by-tensor. 3.8 billion parameters, your GPU, your browser — every glow is a live activation read back from WebGPU. Zero server, zero API key.",
      "category": "Visualization",
      "badge": "applied",
      "githubRepo": "https://github.com/abgnydn/neuropulse",
      "stats": [
        {
          "value": "Phi-3-mini",
          "label": "params rendered live"
        },
        {
          "value": "every tensor",
          "label": "rendered 1:1"
        },
        {
          "value": "offline",
          "label": "zero server calls"
        }
      ],
      "cta": "Watch it think →"
    },
    "barisgunaydin": {
      "key": "barisgunaydin",
      "url": "https://barisgunaydin.com",
      "domain": "barisgunaydin.com",
      "name": "barisgunaydin.com",
      "tagline": "Full-stack developer and researcher — kernel fusion, GPU compute, browser-native ML.",
      "shortDesc": "Personal site and project hub.",
      "category": "Personal",
      "badge": "personal",
      "cta": "About →"
    },
    "markview": {
      "key": "markview",
      "url": "https://markview.ai",
      "domain": "markview.ai",
      "name": "markview.ai",
      "tagline": "Markdown editor.",
      "shortDesc": "Markdown editor and desktop app.",
      "category": "Utility",
      "githubRepo": "https://github.com/abgnydn/markview",
      "cta": "Open →"
    },
    "safenpm": {
      "key": "safenpm",
      "url": "https://safenpm.dev",
      "domain": "safenpm.dev",
      "name": "safenpm.dev",
      "tagline": "npm install, but you see what you're installing.",
      "shortDesc": "Inspect npm packages before adding them to your project — dependency tree, post-install scripts, license, maintainer, file list. Pre-install sanity check for the registry.",
      "category": "Tooling",
      "githubRepo": "https://github.com/abgnydn/safenpm",
      "cta": "Inspect a package →"
    },
    "fusedlora": {
      "key": "fusedlora",
      "url": "https://fused-lora.pages.dev",
      "domain": "fused-lora.pages.dev",
      "name": "fused-lora",
      "tagline": "LoRA fine-tuning of BitNet b1.58 2B4T, live in a browser tab.",
      "shortDesc": "First public demo of fine-tuning a ternary LLM entirely in the browser via WebGPU. Adapters emit as compact .flora files. Includes a Phi-3 LoRA reference.",
      "category": "LLM training",
      "badge": "research",
      "githubRepo": "https://github.com/abgnydn/fused-lora",
      "cta": "Fine-tune in browser →"
    },
    "webgpuDna": {
      "key": "webgpuDna",
      "url": "https://webgpu-dna.pages.dev",
      "domain": "webgpu-dna.pages.dev",
      "name": "webgpu-dna (secondary)",
      "tagline": "Secondary Cloudflare Pages deploy for webgpudna (preview/experimental surface).",
      "shortDesc": "Separate CF Pages project used alongside the canonical webgpudna.com — kept live intentionally. Not a distinct product; same codebase.",
      "category": "Radiobiology",
      "githubRepo": "https://github.com/abgnydn/webgpu-dna",
      "cta": "Open preview →"
    },
    "webgpuq": {
      "key": "webgpuq",
      "url": "https://webgpu-q.vercel.app",
      "domain": "webgpu-q.vercel.app",
      "name": "webgpu-q",
      "tagline": "Full-featured quantum circuit simulator in a single browser tab.",
      "shortDesc": "Statevector + MPS quantum simulator running on commodity hardware via WebGPU compute. Six-level research ladder from bandwidth-bound statevector through MPS, kernel fusion, WebRTC swarm, IBM hardware cross-verify, to chemistry/VQE. No CUDA, no install.",
      "category": "Quantum",
      "badge": "research",
      "cta": "Open the simulator →"
    }
  },
  "CROSSLINKS": {
    "kernelfusion": [
      "webgpudna",
      "gpubench",
      "zerotvm",
      "neuropulse",
      "webgpuq",
      "barisgunaydin"
    ],
    "webgpudna": [
      "kernelfusion",
      "gpubench",
      "zerotvm",
      "neuropulse",
      "webgpuq",
      "barisgunaydin"
    ],
    "zerotvm": [
      "kernelfusion",
      "webgpudna",
      "gpubench",
      "neuropulse",
      "webgpuq",
      "barisgunaydin"
    ],
    "gpubench": [
      "kernelfusion",
      "webgpudna",
      "zerotvm",
      "neuropulse",
      "webgpuq",
      "barisgunaydin"
    ],
    "neuropulse": [
      "kernelfusion",
      "webgpudna",
      "zerotvm",
      "gpubench",
      "webgpuq",
      "barisgunaydin"
    ],
    "barisgunaydin": [
      "kernelfusion",
      "webgpudna",
      "gpubench",
      "zerotvm",
      "neuropulse",
      "webgpuq",
      "markview",
      "safenpm"
    ],
    "markview": [
      "barisgunaydin",
      "kernelfusion"
    ],
    "safenpm": [
      "barisgunaydin",
      "kernelfusion"
    ],
    "fusedlora": [
      "kernelfusion",
      "zerotvm",
      "barisgunaydin"
    ],
    "webgpuq": [
      "kernelfusion",
      "webgpudna",
      "gpubench",
      "zerotvm",
      "neuropulse",
      "barisgunaydin"
    ],
    "webgpuDna": [
      "webgpudna"
    ]
  },
  "AUTHOR": {
    "name": "Ahmet Baris Gunaydin",
    "email": "hi@barisgunaydin.com",
    "github": "https://github.com/abgnydn",
    "linkedin": "https://www.linkedin.com/in/abgnydn/"
  },
  "SAME_AS": [
    "https://barisgunaydin.com",
    "https://kernelfusion.dev",
    "https://gpubench.dev",
    "https://zerotvm.com",
    "https://webgpudna.com",
    "https://neuropulse.live",
    "https://webgpu-q.vercel.app",
    "https://markview.ai",
    "https://safenpm.dev",
    "https://fused-lora.pages.dev",
    "https://github.com/abgnydn",
    "https://www.linkedin.com/in/abgnydn/"
  ],
  "_source": true,
  "_generated": {
    "by": "sites-shared/bin/build-sites.mjs",
    "warning": "DO NOT EDIT. Numbers come from facts.json. Editing this file is how three contradictory copies ended up live."
  }
}