Dogukan Tuna

LatestWorklogSeptember 22, 2026

Worklog: a recurrent looped transformer in Lean

One day, one file, fifty theorems about the Recurrent Looped Transformer.

Read the post
OCT2025NOVDECJAN2026FEBMARAPRMAYJUNJULAUGSEPOCTTODAYWorklog: a recurrent looped transformer in Lean, September 22, 2026Looped transformer in LeanIntroducing cu-Jev (CUDA-Jev), September 20, 2026cu-JevMulti-agent orchestration with Codex, borrowing Fable 5.1 and Fugu Ultra v1.1, September 7, 2026Codex × Claude × FuguVerified Replay Distillation (VRD) recipe for continual learning in verifiable domains, April 12, 2026Verified Replay Distillationautoresearch-mamba: Karpathy-Style Autoresearch for Mamba-2, Mamba-3, and Hybrid Mamba–Transformer MoE, March 25, 2026autoresearch-mambaA self-improving skill catalog for AI agents, March 11, 2026Self-improving skill catalogMem-RLM: Memory-Augmented Inference for Recursive Language Models, February 23, 2026Mem-RLMClaude Code-Time Skill Acquisition with Agent Teams, February 7, 2026Agent teams acquire skillsOn Compression, Computation and the Space Between, February 1, 2026Compression and computationDefeating Nondeterminism in LLM Inference: Reproducing Batch-Invariant Ops (RMSNorm & Tiled Matrix Multiplication) in JAX, November 25, 2025Batch-invariant JAXStreaming Deep Agents and Task Delegation with Real-Time Output, October 20, 2025Streaming deep agents

Every published post on one time axis. Select a dot to jump to it.

Here I aim to write about world and science simulations for semiconductor manufacturing, computational lithography, wafer-scale AI chips and the hardware and software that are designed together, along with the GPU kernels, agents and research tools I build on the way. 11 posts and 22,100 words so far; my repos live on GitHub.

Writing

All 11 posts →

  1. Worklog: a recurrent looped transformer in Lean

    WorklogLean 4formal verificationautoformalization2 min

  2. Introducing cu-Jev (CUDA-Jev)

    Build logCUDASystem OneJev12 min

  3. Multi-agent orchestration with Codex, borrowing Fable 5.1 and Fugu Ultra v1.1

    Build logCodexClaude CodeMCP4 min

  4. Verified Replay Distillation (VRD) recipe for continual learning in verifiable domains

    Researchcontinual learningreinforcement learningLLM post-training5 min

  5. autoresearch-mamba: Karpathy-Style Autoresearch for Mamba-2, Mamba-3, and Hybrid Mamba–Transformer MoE

    Build logMambaMLXAI agents7 min

  6. A self-improving skill catalog for AI agents

    Build logGPU programmingTPUHPC6 min

  7. Mem-RLM: Memory-Augmented Inference for Recursive Language Models

    ResearchLLM inferencememory augmentationrecursive language models7 min

Projects

Built on the side, mostly open source.

Open source

cu-Jev

CUDA-native System One decision engine with a Jev-compatible API.

CUDASystem OneQwen3.5
GitHub

Topics

One row per field. Select a term to filter the writing.

Silicon

,,,,,

Models

,,,,,

Agents

,,,,,

Learning

,,,,,

Theory

,,,,,

About

About →

Dogukan Tuna

Hi, I work on the part of AI that happens a few nanometers from the wafer: world and science simulations for semiconductor manufacturing, computational lithography, wafer-scale AI chips, and hardware and software that co-design together. I post longer writing and worklogs here. My open-source projects are on GitHub.

> AI a few nanometers from the wafer.

>> World and science simulations for semiconductor manufacturing.

>>> Computational lithography, wafer-scale AI chips, HW/SW co-design.

GitHubXMediumGoodreadsRSS