Skip to content

Releases: chsushen/novascientist

Release list

NovaScientist v2.3.0 — Evidence-First AI Research Orchestration & Reproducibility Infrastructure

NovaScientist v2.3.0 — Evidence-First AI Research Orchestration & Reproducibility Infrastructure Latest
Latest

Choose a tag to compare

View all tags
@chsushen chsushen released this 05 Sep 17:04
v2.3.0
f63e8f0

Contract-driven research orchestration, evidence-grounded literature workflows, multi-seed experimentation, statistical evaluation, provenance tracking, reproducible publication artifacts, FastAPI/Streamlit infrastructure, security controls, and automated testing.
NovaScientist is an experimental research-automation and research-infrastructure prototype. Generated hypotheses, empirical interpretations, and publication drafts require human scientific verification.

Assets 2
Loading

NovaScientist v2.0: Autonomous Research-to-Publication Engine

NovaScientist v2.0: Autonomous Research-to-Publication Engine

Choose a tag to compare

View all tags
@chsushen chsushen released this 03 Sep 18:50
v2.0.0
0345be2

NovaScientist v2.0 🔬

We are proud to release NovaScientist v2.0, an interactive conversational research agent and genuine hardware-benchmarking suite that transforms scientific problem formulations into empirical evaluations and publication-ready IEEE Transactions manuscripts.

🚀 Key Highlights & Architectural Features

  • 🌐 Live Web Application: Fully interactive 4-stage studio deployed on Streamlit Cloud at novascientist-cqhrr8wptwmrzjksbr8pyw.streamlit.app.
  • 🧠 Human-in-the-Loop Theory Gate: Dynamic scoping assistant that formulates and halts for user review of formal mathematical proofs (Lemma 1, Theorem 1, Theorem 2) prior to optimization.
  • ⚡ Physical Hardware Benchmarking: Auto-detects Apple Silicon (MPS), NVIDIA (CUDA), or host CPU; runs true multi-seed ($k=5$) PyTorch training loops with model weight checkpointing (.pt).
  • 🛡️ AST Dataflow Integrity Gate: Statically audits execution pipelines via Python's Abstract Syntax Tree to enforce zero train/test contamination.
  • 📚 Verified Literature Retrieval: Live querying via CrossRef and OpenAlex APIs delivering 100% active, non-hallucinated DOIs.
  • 📊 Publication-Grade Visualizations: Automated generation of 5 vector graphics (.pdf and 300 DPI .png) including Architecture Flow, Convergence Curves, Multi-Objective Pareto Frontier, Module Ablations, and 2D Sensitivity Heatmaps.
  • 📄 Compilable IEEE Typesetting: Produces an 8–12 page camera-ready double-column IEEE Transactions PDF and an Overleaf-ready ZIP bundle (main.tex, references.bib, IEEEtran.cls).
  • ✅ Test Coverage: 31/31 unit and integration tests passing with complete offline fallbacks.

📦 Quickstart

git clone [https://github.com/chsushen/novascientist.git](https://github.com/chsushen/novascientist.git)
cd novascientist
python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
streamlit run app.py
Loading