sst autoresearch.stub

---

id: sst-autoresearch name: SST-AutoResearch repo_path: ~/Documents/sw30labs/repos/sst-autoresearch remote_url: https://github.com/sw30labs/sst-autoresearch.git primary_language: Python framework: LangGraph ingested: 2026-04-11 last_commit_date: 2026-04-11 category: agentic-frameworks stacks: [agentic, langgraph]

SST-AutoResearch

Purpose

Speaker State Trajectory analysis via Karpathy-style autoresearch loop. Treats a speaker's voice as a nonlinear dynamical system — extracts acoustic features, embeds into phase space (Takens' theorem), analyzes trajectory for attractors, bifurcations, Lyapunov exponents. LLM drives the research cycle: hypothesize → design → execute → evaluate → reflect → loop.

Architecture

• LangGraph StateGraph (src/graph.py) with 6 nodes
• Acoustic extraction via Parselmouth + librosa
• Dynamics analysis (RQA, Lyapunov, entropy, regime detection)
• Backend support: MLX (Qwen3.5-122B) or Ollama
• Karpathy-style autoresearch loop driven by LLM

Key Features

• Extracts acoustic features from speaker voice
• Embeds trajectory into phase space using Takens' theorem
• Analyzes nonlinear dynamics: attractors, bifurcations, Lyapunov exponents
• LLM orchestrates research cycle: hypothesis → design → execute → evaluate → reflect
• Supports local inference backends (MLX, Ollama)
• Produces visualizations with matplotlib/plotly
• Vector storage with ChromaDB

Dependencies

• langgraph
• langchain
• langchain-community
• librosa
• praat-parselmouth
• numpy
• scipy
• matplotlib
• plotly
• soundfile
• chromadb
• mlx-lm

Key Claims

• Implements a Karpathy-style autoresearch loop where the LLM drives hypothesis → design → execute → evaluate → reflect cycles over speaker voice dynamics (Takens' embedding, Lyapunov exponents, recurrence analysis).

Tags

• python
• langgraph
• mlx
• local-inference
• audio

Cross-references

• sst-autoresearch (shared audio/speech analysis domain)