Skip to content
View MACCCTK's full-sized avatar
🎯
Focusing
🎯
Focusing
4 followers · 5 following

Highlights

  • Pro

Block or report MACCCTK

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
MACCCTK/README.md

Hi, I'm Carlos

I work on AI systems infrastructure: runtime systems, serving infrastructure, agent control planes, and memory-oriented systems for intelligent applications.

My interests sit where AI workloads meet systems engineering: making model-driven systems faster, more reliable, easier to coordinate, and easier to reason about under real implementation constraints.

Current Focus

  • AI runtime and serving: CUDA, profiling, scheduling, memory management, vLLM, SGLang, Triton.
  • Agent control planes: task scheduling, state consistency, recovery, auditability, runtime interfaces.
  • Memory and graph intelligence: structured memory, temporal graphs, GraphRAG, provenance, relationship reasoning.
  • World models and RL: environment modeling, long-horizon planning, policy learning, agentic behavior.
  • Personal AI OS: knowledge-work systems for research, engineering practice, and local automation.

Selected Work

  • coro: A C++20 coroutine runtime exploration for Linux async I/O, focused on coroutine primitives, scheduler design, and the practical engineering needed to make high-performance systems testable.
  • raft: A strongly consistent control-plane project for agent runtimes and long-running task orchestration, exploring how leases, ownership, checkpoints, and idempotency can make distributed agent work safer.
  • sptem-graph: A local-first memory control-plane project for AI agents, focused on auditable belief state, evidence trails, timeline branching, replay, and memory governance.

Some of these projects are still being shaped before public release, so I keep public descriptions high-level and use the work mainly to sharpen system design, implementation discipline, and research taste.

How I Work

  • I prefer small, verifiable changes over broad rewrites.
  • I care about clear documentation, reproducible commands, and scoped commits.
  • I like systems where the architecture can be explained from the code and tested locally.
  • I try to separate product shape, architecture, and task tracking so projects stay understandable as they evolve.
  • I use AI tools as engineering collaborators, but I still treat source code, tests, and docs as the final source of truth.

Contact

Pinned Loading

  1. agent-market agent-market Public

    Python

  2. vllm vllm Public

    Forked from vllm-project/vllm

    A high-throughput and memory-efficient inference and serving engine for LLMs

    Python