Teo Wiki RESEARCH MAP
Article Research Map Public-safe version

Teo Research Map

A public map of how Teo’s projects, methods, and open questions connect

This map is a public-facing layer of Teo Wiki. It is not a full private research notebook; instead, it explains the visible structure behind Teo’s work: visual editing, 3D/scene understanding, video understanding, audio-visual learning, and practical AI systems.

Map thesis. Teo’s projects can be read as attempts to make multimodal AI systems more useful in the real world: edit visual content realistically, understand scenes and videos, compare signals over time, and evaluate systems under practical constraints.
Core identity

Researcher-builder for practical multimodal AI

Connecting computer vision research with systems that can be tested, deployed, and explained.

Visual Editing

Diffusion-based image manipulation

  • Identity-preserving editing
  • In-the-wild robustness
  • AnyBald and realistic hair removal
Scene Understanding

Language-guided 3D reasoning

  • Object-centric grouping
  • 3D scene representations
  • Gaussian Splatting and CLIP-style semantics
Video Understanding

Reference-guided comparison over time

  • Video alignment
  • Industrial anomaly detection
  • RG-VAD and automated display inspection
Audio-Visual Learning

Multimodal temporal modeling

  • Vision + audio representation
  • Temporal alignment and evaluation
  • Future-facing multimodal frameworks

Practical AI applications

The map is held together by a practical systems orientation. Projects are not only demonstrations of model capability; they ask what must be true for a model to work under real conditions: noisy input, long videos, human interaction, industrial constraints, and ambiguous evaluation.

Robotics / HRI

Vision-based Gomoku AI robot

A physical interaction project combining perception, decision-making, and robotic control.

Creative tooling

Cheese! Generative Editing

A creative application direction that tests how generative editing can become usable tooling.

LLM Wiki maintenance loop

This page is also a template for how Teo Wiki can be maintained over time using Karpathy’s LLM Wiki pattern. Public pages should be curated outputs from a private markdown knowledge base, not one-off static copy.

  1. Ingest: add public-safe papers, project pages, reading notes, and research statements into a private raw source folder.
  2. Compile: let the LLM update concept pages such as visual editing, video anomaly detection, and audio-visual learning.
  3. Cross-link: connect projects to concepts, concepts to open questions, and questions back to publications.
  4. Curate: export only public-safe summaries to this website.
  5. Lint: periodically check for stale claims, broken links, weak citations, and missing pages.

Questions this map should keep answering

  • Which research direction best explains Teo’s transition from visual editing to broader multimodal learning?
  • How do practical systems projects shape the kinds of model evaluation Teo cares about?
  • Which concepts should become standalone public wiki pages as the research story matures?