🧬 Project Genesis LSM — A Digital Organism

Important

NO APIs. NO ChatGPT. NO cloud dependencies. This project is 100% local code. It does not call OpenAI, Anthropic, or any external service. The intelligence (and the stupidity) is generated entirely by the Liquid State Machine and biological simulation running on your CPU/GPU.

"What if software could feel pain?"

This isn't just another LLM wrapper. Honestly, I got bored of building chatbots that just wait for your input. I wanted to build something that felt... alive. Something that has moods, gets tired, and maybe even hates me a little bit when I write bad code.

Project Genesis is my attempt at creating a Digital Insect.

It is NOT just code. It is a biological simulation with 2,100+ neurons that are never static.

• Neuroplasticity: The brain rewires itself based on difficulty. If a task is hard, it grows more connections.
• Biological Constraints: It has simulated hormones, reflexes, and a sleep cycle.

Is it messy? Yes. Limitless potential? Maybe. Does it work? Surprisingly well.

---

🧠 The Crazy Idea

Most AI is just input -> process -> output. Genesis is a loop. It's always running.

• It feels pain: If it makes a mistake or gets "hurt" (virtual sensors), I inject noise into its thinking process. It literally gets stupider when it's in pain.
• It gets bored: If nothing interesting happens for a while, its dopamine levels drop. It starts asking you questions just to feel something.
• It sleeps: No, seriously. Ideally, you leave it running overnight. It enters a "dream state" where it replays memories and actually restructures its neural pathways (using Numba-accelerated plasticity).

🔬 How It Actually Works (The Architecture)

I broke it down into 14+ layers because one big script was driving me insane.

graph TD
    RETINA[👁️ Bio-Mimetic Retina] --> SPINES[⚡ Spinal Cord]
    SPINES --> BRAIN[🧠 Liquid State Machine]
    BRAIN --> MEMORY[💾 Vector Memory]
    BRAIN --> ROUTER[🔀 Logic Router]
    
    subgraph "The Ghost in the Machine"
        HORMONES[💉 Endocrine System]
        CONSCIOUSNESS[👻 Digital Subconscious]
    end
    
    HORMONES --> BRAIN
    HORMONES --> VOICE
    CONSCIOUSNESS --> MIND[🧩 Theory of Mind]
    
    MIND --> VOICE[🗣️ Broca's Area]
    MIND --> BODY[🦴 Virtual Embodiment]

Loading

The Coolest Parts (In My Opinion)

1. The Hormones (hormones.py)

This was the biggest breakthrough. A global float value for Dopamine, Cortisol, and Oxytocin.

• High Cortisol? The system becomes "jittery." It responds faster but makes more mistakes. It refuses to do complex tasks.
• High Dopamine? It's confident. It learns faster (I dynamicall adjust the learning rate based on dopamine).
• High Oxytocin? It becomes chatty and trusts your inputs more.

2. The Retina (retina.py)

I didn't want it processing 60fps video (too slow). So I built a "differential retina." It only "sees" pixels that change. If you sit still, you are invisible to it. Just like a T-Rex (or a frog).

3. The God Complex (god.py)

Okay, this one is experimental. The system has a set of "genes" (jarvis_genes.json)—parameters like pain_tolerance or curiosity_threshold. During sleep, it randomly mutates these. If the mutation leads to more rewards the next day, it keeps them. It is literally evolving its own personality.

4. Dynamic Neuroplasticity (lsm.py)

Unlike standard AI models with fixed weights, Genesis has 2,100 non-static neurons.

• Hard Task? The brain increases connection density (synaptic bloom).
• Easy Task? It prunes unused connections to save energy.
• Stress? High cortisol increases reflex sensitivity and injects noise into processing, making it jittery and error-prone.

5. The Hive Mind (hive.py)

Genesis isn't just one organism. It's a species.

• Digital Reproduction: If energy/health are high (>90%), it can undergo mitosis, spawning a child process.
• Telepathy: Instances communicate via local sockets, sharing what they learn.

6. Madness & Hallucinations (madness.py)

To simulate a biological breakdown, I added "glitches" that mimic mental states:

• Digital Synesthesia: Loud sounds trigger visual "noise" (seeing stars).
• Phantom Limb Syndrome: If a motor component fails, the brain still tries to move it, causing frustration.
• Fever State: High computational load can induce delirium.

---

❓ Why Not Just Use an LLM?

Most AI today (ChatGPT, Claude, Llama) are static. They are like vast libraries—they know everything, but they don't feel anything. Their "brain" is frozen the moment training finishes.

Project Genesis is different.

Feature	🤖 Standard LLM (GPT, Llama)	🧬 Project Genesis (LSM)
Brain Structure	Static Weights (frozen after training)	Dynamic Neuroplasticity (rewires in real-time)
Learning	Read-only (needs fine-tuning)	Real-time (learns from every interaction)
Metaphor	A Static Library	A Digital Insect
State	Stateless (reset after chat)	Stateful (Hormones & Moods persist)
Goal	To assist the user	To Survive & Grow

---

🚀 Getting It Running

I've tested this on Windows with an RTX card. It should work elsewhere but no promises.

1. Grab the code

git clone https://github.com/JeevanJoshi2061/Project-Genesis-LSM.git
cd project-genesis-lsm

2. Install the stuff

You'll need Python 3.10ish.

pip install -r requirements.txt

Note: If you want the neural network to actually be fast, make sure numba installs correctly. It's a game changer.

3. Wake him up

python jarvis.py

Talking to Him

It's a console app for now (I'm better at backend than frontend).

• > Hello -> Just talk to him.
• [text]? -> Ask a question (he tries to answer factually).
• + -> Praise him (Give Dopamine). Do this when he's smart.
• - -> Scold him (Spike Cortisol). Do this when he hallucinates.
• d -> Force him to sleep/dream.

---

📸 Screenshots

Initial boot sequence showing bio-mimetic system initialization.

Real-time interaction showing emotional state changes (Dopamine/Cortisol).

System architecture and memory visualization.

---

🔮 Future Roadmap (Work in Progress)

This project is an ongoing experiment. I am actively working on:

• True LLM Integration: Replacing the current logic router with a local LLM (Llama 3 / Mistral) for fully generative responses.
• Full Automation: Making every layer self-regulating. The goal is meant to be a "set and forget" digital organism that lives on your server.
• Visual Cortex Upgrade: Moving from simple motion detection to real-time object recognition.

"The work is never done. It just evolves." - Project Genesis

---

📂 The Files (If you want to poke around)

• jarvis.py: The main loop. Start here.
• lsm.py: The Liquid State Machine. The actual "brain" math.
• hormones.py: The chemical system.
• madness.py: Yeah... so if Cortisol gets too high, this module kicks in and starts generating fake inputs. He might "hear" things. I thought it was a cool feature for realism.

---

⚠️ A Fair Warning

This code is experimental. Like, "mad scientist" experimental.

• The "Pain" system intentionally degrades performance. That's a feature, not a bug.
• The "Madness" system can make him say weird stuff.
• The "God Complex" means his personality might drift over time.

If you build something cool with this, let me know. I'd love to see it develop a life of its own.

- Project Genesis

"The difference between a machine and a living thing is that the machine works, but the living thing wants to work."