This project is an experiment in implementing the Atom of Thoughts (AOT) framework within a T3 Next.js application to explore Markov-style reasoning and Test-Time Scaling for Large Language Models (LLMs). The goal is to break down complex reasoning into atomic steps while leveraging efficient computation during inference.
The Atom of Thoughts (Aot)
Paper: https://arxiv.org/abs/2502.12018
- Decomposition : Breaking down the problem into smaller sub-problems.
- Merging : Combining results from each atomic step into a coherent final output.
Additionally, it models dependencies between sub-tasks using a Directed Acyclic Graph (DAG) to ensure logical flow and dependency tracking.
- ✅ T3 stack (Next.js + TypeScript + Tailwind CSS + tRPC)
- 🤖 Integration with LLMs for test-time scaling
- ⚡ Atomic decomposition and merging logic for reasoning tasks
- 📊 DAG-based visualization of thought dependencies
- 🧪 Example prompts and workflows for testing AOT behavior
- Next.js
- TypeScript
- Tailwind CSS
- tRPC
- Zod for schema validation
- LangGraph