Skip to content
Portfoliolab.

Working systems, taken apart.

Two pipelines I actually build — robot autonomy and language models — rebuilt as things you can drive. Every simulation runs live in your browser. Nothing here is recorded, faked, or calling an API.

Autonomy

Sense → map → plan → act. The pipeline behind a mobile robot.

  1. 01Sense

    LiDAR Playground

    A rotating laser scanner, ray-cast live against a warehouse floor.

    • Ray casting
    • Incidence-angle intensity
  2. 02Map

    Occupancy Mapping

    Turn raw scans into a probabilistic map, one beam at a time.

    • Log-odds occupancy
    • Inverse sensor model
  3. 03Plan

    Path Planning

    Watch A* search a map it discovered for itself.

    • A* search
    • Heuristic weighting
  4. 04Act

    Dynamic Replanning

    Block the robot's path and watch it notice, then find another way.

    • Costmap fusion
    • Line-of-sight discovery

Language

How text actually becomes something a model can compute over.

  1. 05Tokenize

    Tokenizer

    See exactly how a language model chops your text up.

    • Byte-pair encoding
    • Vocabulary merges
  2. 06Retrieve

    Semantic Retrieval

    Search this portfolio by meaning, with a real model running in your tab.

    • Sentence embeddings
    • Cosine similarity

Everything renders client-side on canvas or runs in-tab — no WebGL, no recorded data, no API calls. Source lives in the portfolio repository.