Skip to content
jaeundaPublic

About

No description, website, or topics provided.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Latest commit

 

History

26 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 

Repository files navigation

Deterministic cores for nondeterministic systems.
Tools that make concurrent, AI-assisted, and parallel systems replayable and measurable.


weavegate

Developer tool that automates database race testing in Spring Boot CI

  • Replay · saves the interleaving between sync-points and reruns it against real MySQL.
  • Judge · SQL invariants decide pass or fail, so the verdict does not depend on timing.
  • Prove · on one saved schedule, the vulnerable Spring Boot app fails and the fixed one passes, 20/20 each.
  • Gate · one workflow step, published on GitHub Marketplace, fails the pull request with the schedule, trace, and violating rows attached.

Deterministic replay Go MySQL 8 / InnoDB Testcontainers GitHub Actions

WeaveTrail

Deterministic trade-case replay with measured selection of its column-mapping model

  • Propose · a model maps each column of an unfamiliar trade file to a canonical field, or returns REVIEW_REQUIRED.
  • Check · a closed contract and a dry run reject invalid proposals; a person approves the exact proposal, bound to its hash.
  • Compute · the replay engine re-derives events from stored rows and tests a versioned pattern rule; each finding traces to its rawRowHash.
  • Select · candidates run on a sealed held-out set beside a non-model baseline; six criteria fixed before the run decide whether any may propose mappings.

AI trust boundaries Model evaluation TypeScript Next.js


Performance studies What it measures
surveillance-frontier Time-budgeted stress testing of market-surveillance detection policies on CPU and GPU. The incremental exact method recomputes only the windows a test event touches, cutting a trial from 1,180.55 ms to 3.36 ms (349×, one week of 1-second BTCUSDT bars, one thread), with 0 mismatches against full recomputation across 2,112 checks (C++ / OpenMP / CUDA)
drone-doa-gpu Real-time acoustic direction finding for drones on an 8-mic array. On a Tesla T4, moving the whole pipeline to the GPU cuts 99.4 ms to 2.58 ms per 64-frame batch (Python / CUDA)

Team projects · built with teams across embedded, backend, and AI
Project What it does
Ongi Device Connected medication-dispenser firmware, from caregiver schedule to confirmed intake. The device runs the server's schedule on synced time, opens the right slot, and reports whether the pill was actually taken (C / ESP-IDF)
Team-po / Server Team matching and collaboration for developer side projects. Matching runs in a locked transaction so no one lands on two teams, and each team gets a group chat, GitHub integration, and an AI-generated development guide (Java / Spring Boot)
CarryCheck Hybrid RAG agent for airline baggage rules, with deterministic decisions and guardrailed generation. On a 10-question set, BM25 and Qwen3 embeddings fused by RRF reach Recall@3 1.00, a harness that rejects changed statuses or unknown source IDs passes 10/10, and selective context cuts one request's tokens by 38.6% (Python / FuriosaAI)
Foundations · coursework and labs in kernels, file systems, storage, compilers, and cloud platforms
Project What it involved
xv6-kernel-extensions Kernel changes for saved file versions, observable physical memory, and weighted scheduling in xv6. Copy-on-write snapshots allocate 98.85% fewer extra blocks than a full copy (3 vs 262), and stride scheduling holds 1:2:4 CPU shares within 0.14 points, against 23.81 for round-robin (C)
linux-filesystem-tools File organization tools from interactive copy to a background daemon, plus a read-only ext2 image inspector. Organizing 512 files / 128 MiB takes a median 0.122 s, with all 2,560 copies matching by SHA-256 (C)
storage-io-lab Flash I/O, a hybrid flash translation layer, and fixed-slot record storage on a file-backed emulator. Reusing 250 deleted slots keeps a 1,000-record file at zero growth, and write amplification measures 1.000× to 3.988× by workload (C)
compiler Eight steps from an expression calculator to a C-subset compiler. RD and LR parsing, scoped symbol tables, and type annotation lead to stack-machine assembly that the included interpreter runs (C)
cloud-native-labs Concept notes and hands-on labs on how VMs, networking, and shared storage hold up a container platform. Covers Kubernetes workloads, Services and Ingress, persistent volumes, scheduling and rollouts, CI/CD, and AWS EKS (Kubernetes / AWS)

About

No description, website, or topics provided.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors