Skip to content
View UniverseScripts's full-sized avatar

Block or report UniverseScripts

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
UniverseScripts/README.md

Yoshio Nomura | Software Engineer & Edge Infrastructure

Architecting fault-tolerant, horizontally scaled control planes. My engineering focus bridges the mathematical reality of Machine Learning Operations (MLOps) with the financial rigor of distributed B2B systems. I transition localized, thermal-bound compute models into verifiable, globally routable production environments.

⚙️ Core Architecture & Operational Stack

  • Orchestration & DevOps: Kubernetes (K3s), Docker, GitHub Actions (CI/CD)
  • State & Telemetry: PostgreSQL (Asyncpg), Redis (Lua Pipelines), Prometheus, Grafana
  • Inference & ML: PyTorch, Hugging Face Ecosystem, PEFT (LoRA), RLHF (PPO)
  • Routing & Client: FastAPI, Next.js, TypeScript

🏗️ Active Engineering & Deployments

System: Enterprise Edge Cluster (Distributed B2B LLMOps)

  • Architecture: Kubernetes (K3s) Control Plane with Merchant of Record (MoR) Perimeter.
  • Implementation: Orchestrated a fault-tolerant edge node serving headless inference matrices. Engineered a trans-continental financial perimeter using asynchronous Redis token buckets and PostgreSQL ledgers to strictly decouple stateless routing from stateful billing. Automated the trans-continental deployment matrix using GitHub Actions CI/CD with ephemeral openAPI syntax auditing.
  • Telemetry: Declarative observability matrix deployed via Prometheus and Grafana to mathematically prove sub-millisecond B2B routing latency.

System: Generative AI Optimization & Alignment (FLAN-T5)

  • Architecture: Parameter-Efficient Fine-Tuning & Reinforcement Learning pipeline.
  • Implementation: Applied LoRA adapters to freeze 98.6% of parameters, radically reducing GPU VRAM overhead. Engineered an RLHF pipeline using Proximal Policy Optimization (PPO) and Meta's RoBERTa to mathematically detoxify model outputs while controlling model drift via KL-divergence penalties.

System: Fitnest Infrastructure

  • Architecture: Containerized Microservices Topology.
  • Implementation: Next.js client communicating with a strictly typed FastAPI backend. Engineered real-time WebSocket (wss) routing and automated PostgreSQL schema normalization.
  • Status: Actively deployed via Render.

📡 Telemetry & Documentation

Engineering scale requires distributed communication. I document system designs, debugging protocols, and architectural decisions asynchronously.

  • [LinkedIn] - Professional network and enterprise opportunity pipeline.
  • [X] - Real-time technical commentary, architectural transitions, and system telemetry.
  • [YouTube Shorts / TikTok / Instagram] - Condensed technical broadcasts and visual documentation of the DevOps lifecycle.

🔬 Current Research Focus

Actively preparing for rigorous software engineering research environments and analyzing distributed system bottlenecks. Seeking global integration challenges that demand strict adherence to algorithmic efficiency, CI/CD automation, and scalable backend design on constrained edge silicon.

Pinned Loading

  1. nextjs-marketplace-free nextjs-marketplace-free Public

    This is the Open Source UI version of the **Marketplace Starter Kit**.

    TypeScript 1

  2. cs50x_fake_news_detector cs50x_fake_news_detector Public

    Fake News Detector web application that extracts text from a "pdf" file and calls BERT's Fake News Detector model on Hugging Face for inference.

    Python

  3. fitnest-webapp fitnest-webapp Public

    A web application in-development to help Vietnamese students who reside in HCMC to find their appropriate place and partner.

    TypeScript

  4. job-scrapper-ai job-scrapper-ai Public

    Python

  5. llmops llmops Public

    A globally routed, zero-cost edge inference node for quantized LLMs. Bypasses hyperscaler GPU costs utilizing Docker, FastAPI threadpools, in-memory rate limiting, and Cloudflare Zero Trust tunnels.

    Python

  6. local-rag-api local-rag-api Public

    An enterprise-grade, Dockerized RAG (Retrieval-Augmented Generation) API boilerplate. Engineered for immediate deployment in production environments, this backend solution provides the foundational…

    Python