Cognitive AI Research Copilot
A research pipeline for cognitive-distortion identification and rational-response generation.
Original project: Cognivia · Qi Chen, Siria Xiyueyao Luo, Jian Wang, Yuan Shi, Haocong Rao, and Xuejiao Zhao
01 / Overview
The system at a glance.
Cognivia is the official research implementation accompanying a 2026 paper on a CBT-oriented language-model pipeline. The work combines expert seed curation, structured dataset augmentation, task-oriented LoRA fine-tuning, and several evaluation paths.
- Original project
- Cognivia
- Created by
- Qi Chen, Siria Xiyueyao Luo, Jian Wang, Yuan Shi, Haocong Rao, and Xuejiao Zhao
- License
- CC BY-NC 4.0; commercial use requires a separate agreement with the authors
- Source review
- August 22, 2026
High-stakes research boundary
Research is not clinical care.
Cognivia is academic research—not medical advice, diagnosis, treatment, crisis support, or a clinically validated product. Mental-health systems require independent clinical, legal, privacy, security, and safety review before any real-world use.
02 / Challenge
The engineering problem.
Mental-health language systems require exceptionally careful data provenance, boundaries, evaluation, and human oversight. A persuasive model response is not evidence of clinical safety or suitability.
The repository focuses on dataset construction and research evaluation for identifying cognitive distortions and producing structured rational responses.
03 / System design
How the architecture responds.
The research workflow begins with a curated cognitive-triplet seed dataset. It then filters eligible PsyQA questions and generates augmented triplets through staged prompting. A Qwen2.5-7B-Instruct model is fine-tuned with LoRA, followed by NLP, existing-criteria, and CogEval-oriented evaluation scripts.
- 01
Curate expert reference triplets
- 02
Obtain PsyQA through its official access process
- 03
Identify candidate cognitive distortions
- 04
Generate structured rational responses
- 05
Fine-tune with LoRA
- 06
Evaluate with multiple criteria
04 / Technology
The implementation stack.
A concise, repository-backed view of the primary platforms, protocols, models, and runtime tools.
- Python
- Qwen2.5-7B-Instruct
- LoRA
- DeepSeek
- GPT-5 Mini
- OpenAI-compatible APIs
- NLP evaluation
- PsyQA
05 / Highlights
What makes the system notable.
- Three-stage research framework
- Explicit dataset access constraints
- Task-oriented parameter-efficient fine-tuning
- Multiple evaluation scripts
- Published paper and citation record
06 / Repository evidence
Verifiable signals.
These statements are derived from the project’s current README, source structure, or license—not from NexLoomix client work.
- The repository reports selecting 9,437 samples from approximately 22,000 PsyQA entries through a prompt-based filtering stage.
- The documented fine-tuning target is Qwen2.5-7B-Instruct using LoRA.
- PsyQA cannot be redistributed by this repository; users must obtain it through the official source and usage process.
07 / NexLoomix perspective
The transferable product lesson.
For any high-stakes AI product, the transferable lesson is governance: provenance, access rights, evaluation design, boundary language, and qualified human review must be part of the architecture.
This interpretation is editorial analysis by NexLoomix. It is intentionally separated from the verified repository evidence above.
08 / Source & attribution
Credit where it belongs.
Research reference only. This page does not claim clinical validation, medical-device status, therapeutic efficacy, or NexLoomix authorship.
- Repository
- Cognivia on GitHub
- Author / organization
- Qi Chen, Siria Xiyueyao Luo, Jian Wang, Yuan Shi, Haocong Rao, and Xuejiao Zhao
- License summary
- CC BY-NC 4.0; commercial use requires a separate agreement with the authors · review license
- NexLoomix relationship
- Independent editorial showcase; no authorship, client relationship, partnership, or endorsement claimed.
09 / Related NexLoomix services
Build a system around the lesson.
These capabilities connect the reference architecture to a new product shaped around your own users, data, constraints, and ownership.
From reference to product
Build an intelligent system around your business.
Bring us the workflow, constraint, or opportunity. We’ll help shape the right product and architecture.