WaveGrating
A wave-optics renderer for Blender, in development.
Applied intelligence.
Independent AI product engineer, Los Angeles
Selected workSelected work
Modly × Hunyuan3D
An image-to-3D extension with staged GPU memory use.
EARS Bridge
A headphone measurement bridge for Dirac Live.
Local AI Systems
A self-hosted AI model for coding tools.
Hermes Classic Gold
A gold theme and session telemetry for Hermes Desktop.
Hermes Warm Compaction
A context plugin for Hermes Agent.
Open sourcecontributions.
When a tool I use breaks, I trace the fault and send its maintainers a fix or a field report. Of the five pull requests below, two are merged and three are in review.
- Prefix-cache repair for the drafter group
The cache never hit on this recipe. With the fix, the time to first token for a repeated 262,144-token prompt fell from 191.7 s to 2.98 s.
- GB10 unified-memory field report
Documents the load guard, the startup-check trap, and fault gating for the shared memory of a DGX Spark.
- Cut transient memory in NVFP4 Marlin scale-factor computation
Weight loading copied the scales of every expert at once. On a test tensor one eighth of production size, memory growth fell from 1,011 MiB to 3 MiB.
- Separate context usage and preserve acting output budgets
Hermes counted an 8,000-token acting request and a 230,000-token advisor request as one 238,000-token context. The fix measures context from the acting request.
- Rearm the compaction budget after LCM progress
After three compactions in one long turn, automatic compaction stopped. A new test runs five compactions in one turn.
A littleabout me.
I'm Shayan Bianconi, an independent AI product engineer based in Los Angeles.
My work includes model integration, installers, memory management, health checks, and documentation. My background is in audio, hardware, and simulation. I test changes on my own machines and publish measurements with their conditions.
I also maintain Apple HIG for Agents, which helps coding agents use Apple's interface design guidelines.
I'm open to full-time AI product engineering roles.