Skip to content
View oabdrabo's full-sized avatar
2 followers · 2 following

Sponsoring

@vllm-project

Achievements

Achievement: Public SponsorAchievement: QuickdrawAchievement: YOLOAchievement: Pair ExtraordinaireAchievement: Pull Sharkx2

Achievements

Achievement: Public SponsorAchievement: QuickdrawAchievement: YOLOAchievement: Pair ExtraordinaireAchievement: Pull Sharkx2

Organizations

@pyxis3-ai

Block or report oabdrabo

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
oabdrabo/README.md

Hi, I'm Omar 👋

Founder @pyxis3-ai · autonomous AI infrastructure operations · London

ex-Seldon (vLLM, LLM inference) · ex-AWS Industry Specialist (semiconductors, AI/ML) · ex-Dell EMC · ex-IBM · author of the canonical AWS guide on decoupling RDS from Elastic Beanstalk — the same procedure AWS's official YouTube channel cites as "Watch Omar's video to learn more".

I work on production AI and cloud infrastructure: autonomous infrastructure operations, LLM inference serving, and Kubernetes-native operations.

🚀 What I'm building

PYXIS3 — autonomous AI that runs your cloud and data-center infrastructure operations across AWS, Google Cloud, Azure, VMware, Nutanix, and on-prem, spanning cost, capacity, reliability, security, and governance, within your guardrails. One subscription, never a share of savings. Org: @pyxis3-ai.

🛠️ Public projects

All under @pyxis3-ai:

  • pyxis-arch — public architecture notes on model-agnostic LLM serving: design notes, runtime-adapter abstraction, decision rationale.
  • vllm-bench — throughput + latency benchmark for OpenAI-compatible LLM endpoints (vLLM, TGI, llama.cpp, Ollama). Measures TTFT, TPOT, request and token throughput at percentiles. Async; two-dependency footprint. MIT.
  • llm-serving-cookbook — production recipes for K8s-native vLLM-first serving. vLLM-on-EKS, KEDA autoscaling, token economics, TTFT optimisation, runtime selection. Apache-2.0.
  • awesome-model-agnostic-llm — curated list of model-agnostic LLM tooling: serving runtimes, routers, evaluators, observability, standards, open weights. CC0.
  • noor — semantic search over the Quran + Hadith corpus. Arabic-aware multilingual embeddings on sqlite-vec. FastAPI + Vue. Runs as a single Docker image, no external services.
  • lens — in-cluster Kubernetes observability with in-browser kubectl exec. Vue 3 + Bun. Single binary, ServiceAccount-token auth. Built for ML-serving and GPU clusters.

📚 Published

🏛️ Background

  • Seldon Technologies · 2025–present · Senior Solutions Engineer on the production MLOps platform: vLLM-based LLM inference and multi-tenant model serving on Kubernetes.
  • AWS London · 2022–2025 · Solutions Architect, Industry Specialist for the semiconductor industry vertical — AI/ML workloads on Inferentia, Trainium, SageMaker, Bedrock.
  • AWS Cape Town · 2017–2022 · Cloud DevOps Engineer. Authored two AWS Knowledge Center articles + the companion AWS YouTube video.
  • Dell EMC · 2016–2017 · Storage engineering (Isilon).
  • IBM · 2016 · Cloud infrastructure.
  • Earlier · OrecX, HONEST · Egypt, 2007–2012.

🧰 Stack

vLLM · Triton · Kubernetes · KEDA · Helm · Prometheus · Caddy · AWS · GCP · Azure · Python · Go · TypeScript

📫 Reach me

LinkedIn Email

Maintenance

This repository is maintained with small, reviewable updates. Supporting documentation lives in docs/, example inputs live in examples/, and lightweight validation notes live in tests/smoke/.

Pinned Loading

  1. pyxis3-ai/lens pyxis3-ai/lens Public

    AI/LLM-serving observability for Kubernetes. Inspect vLLM/TGI/llama.cpp inference pods, browse resources, kubectl exec in-browser. Single Bun binary, in-cluster SA-token auth.

    Vue 1

  2. pyxis3-ai/noor pyxis3-ai/noor Public

    Multilingual AI semantic search (Arabic + English). Sentence-Transformers embeddings + sqlite-vec vector store + Arabic NLP normalisation. Sub-second on CPU. Quran + Hadith demo corpus.

    HTML 1

  3. pyxis3-ai/vllm-bench pyxis3-ai/vllm-bench Public

    Throughput + latency benchmark for OpenAI-compatible LLM endpoints (vLLM, TGI, llama.cpp, Ollama). TTFT, TPOT, throughput, percentiles. Runtime-agnostic.

    Python 1

  4. pyxis3-ai/awesome-model-agnostic-llm pyxis3-ai/awesome-model-agnostic-llm Public

    Curated list of model-agnostic LLM serving runtimes, routers, evaluators, and standards. Run LLMs without locking into one vendor.

    2 3

  5. pyxis3-ai/llm-serving-cookbook pyxis3-ai/llm-serving-cookbook Public

    Production recipes for running open-source LLMs on Kubernetes. vLLM-first, model-agnostic. vLLM-on-EKS, KEDA autoscaling, token economics, TTFT optimisation, runtime selection.

    2

  6. pyxis3-ai/pyxis-arch pyxis3-ai/pyxis-arch Public

    Public architecture and design notes from PYXIS3 on operating LLM and AI inference workloads on Kubernetes: runtime selection across vLLM/TGI/llama.cpp, GPU utilisation, and fair-share scheduling.

    1