CorbenicAI

CorbenicAI

hyper-efficient infrastructure primitive
#98157664 followers38 following
✨ 3 day streak
>10,000All time
12 KP

About

AI systems engineer and founder working at the inference-infrastructure layer of large language models: deterministic KV-cache reuse, byte-exact deduplication, and verified memory instead of recomputation. Sole author of four arXiv preprints published in 2026, each released with pre-registered methodology, reproducible artifacts, and cryptographic receipts. In parallel I lead enterprise AI delivery at bpost Group, where the same ideas meet real constraints on cost, latency, and governance. I would rather report a number that survives scrutiny than one that reads well

Badges

Gone streaking
Gone streaking
Gone streaking 5
Gone streaking 5

Maker History

Forums

Galahad: The AI That Never Forgetsp/galahad-the-ai-that-never-forgets•

7h ago

AI memory shouldn’t be limited by VRAM

Galahad is out of beta, and we re proud of what we ve built. We re launching on Product Hunt to learn what s missing, hear how it works for different teams, and understand how it affects GPU usage and agent performance.

If you try it, we d love to hear what works for you we open Beta today at 15h belgium time

•

5h ago

Galahad: The AI That Never Forgets - Make Any LLM 20x Faster & 100% Accurate

Galahad installs on the GPU servers you already rent and makes the same hardware do multiples of the work. Every number measured, sha-256 verified.
Aneirinp/aneirin•

3mo ago

use fewer tokens with Claude Code and Copilot, same answers

I made Aneirin, a free local tool that lowers the token cost of Claude

Code and GitHub Copilot. It runs on your own computer, uses your own

account, and removes repeated text the model already saw. You get the

same answers for a smaller bill. Nothing leaves your machine.

View more