Language-driven cognitive architecture for AGI from scratch, alone, meet ALLA

1 darwinSir 1 7/18/2025, 6:09:14 PM dx.doi.org ↗

Comments (1)

darwinSir · 3h ago
Hey HN, I'm a 19-year-old from Indonesia. Over ~3 months, I built ALLA — an experimental, curiosity-driven cognitive architecture designed to simulate emergent general intelligence from scratch.

ALLA doesn't use datasets, pretrained models, embeddings, or LLMs. It learns by asking, understanding, building concepts, and reflecting — just like a child. The system is entirely symbolic, recursive, and modular.

What ALLA does: Learns new concepts through user interaction

Builds its own semantic network (concepts and relations)

Has self-generated goals (e.g., UNDERSTAND, REMEMBER)

Reflects on its memory to form abstractions and analogies

Grows knowledge recursively — from 0 → 1M semantic connections in ~1 hour

ALLA's architecture includes 7 modules: Command Parser, Semantic Memory, Planner, Execution Engine, Goal Manager, Reflection Cycle, and Semantic Bootstrapper.

No Copilot suggestions dictated structure. I guided every logic block, using Copilot only for syntax. This project is 100% solo — from engineering to philosophy.