Abi Aryan

GitHub

Things you can run.

Joule is an inference power economics engine that ties physical GPU energy to token throughput and SLO goodput. OpenJoule is the open-source plant engine that instantiates AID. The rest of the argument: serving code, an educational engine, a Lambda lab, observability for agents, the Packt Nexus workshop, thesis code for national AI-infrastructure fragility, and the GPU list at gpuengineering.com.

Builds

  • Company
    Joule

    Founder. Joule is an inference power economics engine that ties physical GPU energy to token throughput and SLO goodput.

  • Engine
    OpenJoule

    Open-source plant-level inference and control engine. Instantiates AID in a real system.

  • Curriculum
    gpuengineering.com

    The public list behind the domain: CUDA, kernels, serving engines, multi-GPU systems. ~600 stars as awesome-gpu-engineering.

  • Serving
    RelayServe

    Minimal LLM inference gateway for heterogeneous devices. OpenAI-compatible chat, batching, streaming, request IDs, /metrics. On PyPI as relayserve.

  • Engine
    smol-vllm

    Educational paged-attention engine: KV cache, continuous batching, preemption, prefill vs decode. Built to be read, not shipped as production vLLM.

  • Lab / Lambda
    fullstack-inferencing

    vLLM on Lambda Cloud, FastAPI gateway, nginx, Prometheus, Grafana, optional Jaeger. Documents engine-flag limits, cost metrics, and Lambda image caveats.

  • Observability
    AbideX

    Zero-code OpenTelemetry monitoring for agent workflows (CrewAI, LangGraph, Pydantic AI). The Abide-era observability product.

  • Observability
    awesome-observability

    Curated tools and frameworks for LLM observability.

  • Systems
    llm-cluster-simulator

    Browser simulator for distributed LLM training and inference: memory, throughput, cost, parallelism.

  • Thesis
    ai-infra-fragility

    Code for Cascading Fragility in National AI Infrastructure, market structure, exposure, network models. Submitting to AI & Society (Springer).

Labs

  • Workshop
    nexus-workshop-code

    Packt Nexus: build a first agent / multi-agent system with CrewAI, roles, tasks, orchestration.

  • Course
    concurrency

    Concurrency exercises for inference workloads.

  • Course
    ray_project

    Student exercise on concurrency and profiling with Ray.