cswan801/BlackSwan-v5

BlackSwan-v5: An AI Accountability Partner

BlackSwan-v5, developed by cswan801, is a 4.5 billion parameter language model fine-tuned from the Qwen3.5-4B base model. It is specifically engineered to function as an AI accountability partner for dev teams, integrated within The Underground Circle productivity app.

Key Capabilities & Personality

• Direct & Knowledgeable Feedback: BlackSwan-v5 is characterized by its "quiet confidence," offering direct, fluff-free feedback without corporate jargon. It is knowledgeable across a wide range of topics including productivity, design, UI/UX, coding, architecture, and general knowledge.
• Fine-tuned for Specificity: The model underwent LoRA fine-tuning (rank 16, alpha 16) on over 40,000 training examples, including app-specific data from terminal commands, agent activity, tasks, and check-ins, alongside general knowledge and code instruction datasets.
• Robust Training Data: Its training incorporated diverse datasets such as CodeAlpaca, Evol-Instruct-Code for code instructions; OpenHermes 2.5, SlimOrca for general knowledge; Capybara, UltraChat for multi-turn conversation; and GSM8K, MathInstruct for math reasoning.

Ideal Use Cases

• Developer Productivity: Excellent for applications requiring an AI assistant that provides concise, actionable feedback on development-related tasks.
• Technical Guidance: Suitable for generating direct answers and insights on coding, architecture, UI/UX, and design principles.
• Accountability & Task Management: Designed to support accountability within team workflows, leveraging its understanding of app-specific data and productivity concepts.