Skip to content
View clsandoval's full-sized avatar

Block or report clsandoval

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.

Content in all repositories owned by your account will be closed.
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
clsandoval/README.md

carlos sandoval

senior AI engineer. i build agent platforms, and the metering, evaluation and observability that keep them running in production.


work

  • pymc-labs/daimon: i created and own daimon, pymc labs' open-source agent platform for discord and slack. built on anthropic managed agents: five adapters (discord, slack, CLI, MCP, scheduler) over one turn pipeline, with a guild or workspace as the tenant and isolation enforced at the database layer, so a single anthropic key safely serves every install. answers come back in-thread as a fitted model, a chart, and a runnable marimo notebook.
  • nuts-and-bolts-ai: cheerful, an AI creator-marketing platform for US brands. i own the durable-execution layer on temporal, the cost and metering stack, the RAG reply-drafting pipeline and the evaluation around it, and the google CASA security certification.

projects

  • bidkita: search engine for philippine government tenders, 16,000 open notices and an award-relationship graph across 5.08 million records. the AI mode is a server-side agent loop on the pi SDK with typed tools shelling to a tested CLI: zero hallucinated identifiers, ~90% cache hit, ~₱0.22 per turn
  • make-film: a claude code skill that renders a product film from HTML, CSS and GSAP via playwright and ffmpeg. deterministic render, no motion-graphics suite, no diffusion model
  • philippine license plate recognition: detection (yolo) then character recognition (lprnet) then edge deployment (coral tpu). the research behind our TENCON 2022 paper
  • alphazero for connect 4, from scratch: self-play reinforcement learning with MCTS
  • ai swim coach: stroke analysis with computer vision
  • contributed to unifyai/ivy: torch.affine_grid implementation, paddle backend activation fixes

interests

  • agent harness design, and where capability actually comes from in a harness/model pairing
  • LLM cost metering, attribution, and evaluation
  • bayesian inference and media mix modeling (pymc ecosystem)

Popular repositories Loading

  1. LPRnet-keras LPRnet-keras Public

    Philippine license plate recognition network — Keras implementation with custom architecture modifications (Part 2 of 3 in LPH pipeline)

    Jupyter Notebook 4

  2. coral_tpu coral_tpu Public

    edge AI model deployment toolkit for Google Coral TPU — Part 3 of 3 in the Philippine license plate recognition pipeline

    Jupyter Notebook 1

  3. claude-autopilot claude-autopilot Public

    Claude Code plugin for dispatching autonomous work to Claude Managed Agents with structured async brainstorming

    Python 1

  4. cs_21_project cs_21_project Public archive

    CS 21 Course Final Project

    SystemVerilog

  5. ocaml-probset ocaml-probset Public archive

    Problem set for my Ocaml class

    OCaml

  6. object-detection-in-keras object-detection-in-keras Public archive

    Forked from Socret360/object-detection-in-keras

    Python