Grounded in peer-reviewed neuroscience research
See the emotion in your text, and neutralize it.
Novasyne reads the emotional signal in any message, maps it to the brain regions the research ties to each emotion, and rewrites it to read calm and clear without changing what you mean.
Anger, fear, sadness, joy, disgust, surprise and neutral, scored for every message.
Each emotion mapped to the cortical and limbic regions research ties to it.
Flags low-mood language and machine-sounding phrasing alongside the emotion.
The rewrite stays faithful to your original message, calmer wording, same intent.
See it in action
Choose a real message and watch it change, the emotion breakdown, the neutral rewrite, and the brain-region map, side by side.
| Scores | After | Change |
|---|
Emotion to brain-region activation, a top-view glass brain mapped from published neuroscience. Illustrative, not a clinical or diagnostic tool.
Manuscript Assistant
Upload a full LaTeX manuscript and we improve the clarity and flow of each section's prose for publication, keeping your meaning, citations, numbers, and equations intact. Here is an LLM-generated abstract, before and after, section by section.
Developing biologically grounded neural simulators typically requires substantial domain expertise and extensive software engineering, creating barriers to reproducibility and accessibility in computational neuroscience. In this work we introduce a structured large-language-model (LLM)–based workflow for constructing complex scientific simulators. The approach encodes a simulator's mathematical specification, architecture, parameterization, and validation procedures as a reusable prompt “skill” that enables an LLM to generate a complete implementation through interactive development. This framework is intended to provide a generative form of methodological reproducibility in which complex simulation systems can be reconstructed from a formal specification rather than only distributed as static code.
As a demonstration case, we use the LLM workflow to construct a biologically grounded spiking neural network simulator that converts continuous electroencephalography (EEG) recordings into plausible cortical spike dynamics. EEG temporal fluctuations are encoded into neural drive using an ON/OFF population scheme and used to drive a hierarchical network of conductance-based neurons with excitatory–inhibitory connectivity, synaptic plasticity, structural adaptation, and metabolic constraints.
Simulations driven by EEG recordings from 36 subjects under resting and task conditions produce stable and biologically plausible spiking activity with realistic firing statistics and adaptive network structure. A multi-class validation framework evaluates input–output behavior, temporal sensitivity, and dynamical stability across the cohort. These results demonstrate that LLM-guided development can generate complex, mechanistically grounded neural simulators while preserving transparency, validation, and reproducibility.
Building biologically grounded neural simulators typically requires both deep domain expertise and substantial software engineering, and this dual demand limits reproducibility and accessibility in computational neuroscience. We describe a structured workflow that uses large language models (LLMs) to construct complex scientific simulators. In our approach, a simulator's mathematical specification, its architecture, parameterization, and validation procedures are encoded as a reusable prompt “skill,” from which an LLM generates a complete implementation over the course of interactive development. What the framework targets is a generative form of methodological reproducibility: complex simulation systems can be reconstructed from a formal specification instead of being distributed only as static code.
As a demonstration case, we applied the LLM workflow to build a biologically grounded spiking neural network simulator that converts continuous electroencephalography (EEG) recordings into plausible cortical spike dynamics. Temporal fluctuations in the EEG are encoded into neural drive through an ON/OFF population scheme, which then drives a hierarchical network of conductance-based neurons. This network incorporates excitatory-inhibitory connectivity, synaptic plasticity, structural adaptation, and metabolic constraints.
When we drove the simulations with resting and task EEG from 36 subjects, the spiking activity remained stable and biologically plausible, with firing statistics in a realistic range and network structure that adapted over time. Within a multi-class validation framework, we examined input-output behavior across the cohort, along with temporal sensitivity and dynamical stability. The results indicate that LLM-guided development can yield complex neural simulators grounded in mechanism without sacrificing transparency, validation, or reproducibility.
| Section | What changed |
|---|---|
| Framework & motivation | Opening recast in active voice and stock connective phrasing removed. Every term, including the “skill” framing, retained. |
| Methods | A single long sentence split into shorter statements. All method names and acronyms (EEG, ON/OFF, conductance-based) preserved. |
| Results & conclusion | Findings stated in plain narrative order and the closing reworded, with no claim, number, or cohort detail added or dropped. |
How it works
Three steps, on one page, no back-and-forth.
Paste your text
An email, a message, or a draft written by an AI. Up to 500 words.
See the signal
We score seven emotions, flag low-mood and machine-sounding tone, and map the activation onto the brain.
Get the rewrite
A calm, plain version of your text that keeps its meaning, ready to copy and send.
Where people use it
Improve emails
Turn a heated or anxious draft into a measured, professional note before you hit send.
Refine AI-generated text
Strip the over-eager, over-formal tone out of LLM output so it reads plainly and human.
Read emotional dialogue
Quantify the emotion in a message or conversation with clear, per-emotion scores.
Grounded in peer-reviewed research
The emotion model and its brain-region interpretation layer are built on published neuroscience methodology. Every neutralization is scored by validated emotion and mood models, measured, not guesswork.