GeezerVison

`$ whoami`

class LooLo:
    """Researcher · Vibe Coder · Agent Developer"""

    name       = "luoolu"
    alias      = "LooLo — Algorithm Architect & AGI Explorer"
    location   = "Earth (for now)"
    roles      = ["Researcher", "Vibe Coder", "Agent Developer"]
    focus      = ["AGI", "Computer Vision", "RL", "Time-Series", "LLM Agents"]
    philosophy = "The boundary between intelligence and code is dissolving — fast."
    goal       = "Build systems that think, perceive, and autonomously decide."

    currently  = {
        "researching": "Foundation models for sequential decision-making",
        "vibe_coding": "Turning ideas into reality with AI-assisted flow",
        "building":    "Multi-agent autonomous systems with tool use & planning",
    }

    open_to = [
        "research collaborations",
        "frontier AI discussions",
        "agent development partnerships",
        "interesting problems that need creative solutions",
    ]

    def __repr__(self):
        return "A mind obsessed with making machines think — and coding in the vibe."

Roles & Identities

Researcher

Exploring the frontiers of AI

AGI Computer Vision RL Time-Series Foundation Models

Deep research in machine learning, autonomous reasoning, and scientific discovery systems.

Vibe Coder

Coding in the flow state

AI-Assisted Dev Rapid Prototyping Creative Engineering

Harnessing AI tools to transform ideas into working systems at the speed of thought.

Agent Developer

Orchestrating AI swarms

LLM Agents Tool Use MCP Multi-Agent Systems

Building autonomous agents that reason, plan, use tools, and collaborate to solve problems.

Research Domains

Computer Vision

Large-scale self-supervised pre-training (DINOv2, MAE)
Fine-grained instance segmentation (Mask2Former)
3D scene reconstruction & spatial understanding
Geological thin-section automated analysis

Time-Series Forecasting

AutoML ensemble pipelines (AutoGluon v1.2+)
Probabilistic & point forecasting at scale
Bayesian hyperparameter optimization (HPO)
RL-augmented hybrid prediction (KL-8 · v44+)

Reinforcement Learning

Model-based & offline RL for sample efficiency
Policy optimization: PPO / QR-DQN / Meta-Ensemble
Multi-objective & constrained policy learning
Sequential decision-making under uncertainty

Foundation Models & AGI

Vision-Language models & multimodal reasoning
TimesFM, FalconCore, Latent Diffusion Models
Autonomous agent frameworks: tool use & planning
Self-improving & meta-learning architectures

Agent Development

          +---------------------+
          |   Orchestrator LLM  |
          |  (Planning & Reason) |
          +----------+----------+
                     |
        +------------+------------+
        |            |            |
   +----v----+  +----v----+  +----v----+
   | Research |  |  Code   |  | Review  |
   |  Agent   |  |  Agent  |  |  Agent  |
   +----+-----+  +----+----+  +----+----+
        |            |            |
   +----v----+  +----v----+  +----v----+
   |  Web    |  |  File   |  | Security|
   | Search  |  |  System |  |  Scan   |
   |  Tools  |  |  Tools  |  |  Tools  |
   +---------+  +---------+  +---------+

Building multi-agent systems where specialized AI agents collaborate autonomously — each with distinct tools, memory, and reasoning capabilities.

Layer	What I Build	Stack
Orchestration	Multi-agent pipelines with planning & delegation	`Claude Code` `LangGraph` `CrewAI`
Tool Integration	MCP servers, function calling, API orchestration	`MCP Protocol` `Tool Use` `REST/gRPC`
Memory & State	Persistent agent memory, context management	`Vector DB` `FAISS` `Redis`
Evaluation	Agent benchmarks, reliability testing, evals	`Custom Evals` `W&B` `Pytest`

Vibe Coding Philosophy

  Vibe Coding Workflow
  ════════════════════

  1. IDEATE     Dream it up, describe the vision
       ↓
  2. PROMPT     Translate intent into AI instructions
       ↓
  3. GENERATE   Let AI write the first draft
       ↓
  4. CURATE     Review, refine, inject domain expertise
       ↓
  5. SHIP       Deploy and iterate at lightning speed

  Tools: Claude Code · Cursor · GitHub Copilot · v0

Current Focus

┌─────────────────────────────────────────────────────────────────────────────┐
│  ACTIVE RESEARCH & DEVELOPMENT                                              │
│                                                                             │
│  [Research]                                                                  │
│  ▸ RL-enhanced time-series forecasting for complex sequential data          │
│  ▸ Foundation model adaptation for geological computer vision               │
│  ▸ Sparse reward RL in partially-observable real-world environments         │
│                                                                             │
│  [Agent Development]                                                        │
│  ▸ Multi-agent LLM systems for autonomous scientific reasoning              │
│  ▸ Tool-augmented agents with MCP protocol integration                      │
│  ▸ Self-improving agent loops with evaluation-driven iteration              │
│                                                                             │
│  [Vibe Coding]                                                              │
│  ▸ AI-first development workflows & rapid prototyping pipelines             │
│  ▸ Claude Code extensions & custom skill development                        │
└─────────────────────────────────────────────────────────────────────────────┘

Interactive Hub — 01 Digital World

Live demos of AGI concepts and cutting-edge algorithms. Click -> explore -> fork

	Demo	Description	Stack
CV	Machine Vision	DINOv2 + Mask2Former thin-section segmentation	`PyTorch` `CUDA` `MMSeg`
TS	Time-Series	KL-8 sequence prediction · AutoGluon v1.2 + RL v44	`AutoGluon` `Ray` `XGBoost`
RL	Reinforcement Learning	PPO / QR-DQN / Meta-Ensemble sequential agents	`SB3` `Gymnasium` `RLlib`
AGI	AGI Playground	Multi-agent LLM + Vision LLM co-reasoning system	`LangChain` `GPT-4V` `FAISS`

Tech Stack

Languages

ML / DL Frameworks

Agent & AI Tools

Infrastructure & MLOps

A Day in the Life

Morning    Research papers, experiment design, literature review
Afternoon  Vibe coding — prototype ideas with AI assistance
Evening    Agent development — build & test multi-agent pipelines
Night      Open source contributions & community collaboration

GitHub Metrics

Connect

_{"01 Digital World" — Exploring, sharing, and co-building the future AI ecosystem.

If my work sparks an idea or saves you time, a star is the best thanks.}