alphanozcan/essAi-9b

VISIONPricing:Input $0.431 / Cached $0.0862 / Output $1.12Concurrent Unit Cost:1Model Size:9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 1, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

alphanozcan/essAi-9b is a 9 billion parameter Qwen3.5-9B model fine-tuned by alphanozcan specifically for generating authentic college application essays. It excels at producing Common App style personal statements with a natural human voice, specific personal detail, and honest reflection. The model leverages a two-stage fine-tuning process on human-written essays, including SFT and DPO, to achieve its specialized writing style. With a 32768 token context length, it is optimized for generating detailed and nuanced long-form essays.

Loading preview...

Model Overview

alphanozcan/essAi-9b is a specialized 9 billion parameter language model, fine-tuned from Qwen3.5-9B, designed to generate authentic college application essays. It focuses on producing Common App style personal statements that mimic a natural human voice, incorporating specific personal details and honest reflection. This model is the larger counterpart to alphanozcan/essAi (Qwen3-4B).

Training Methodology

The model underwent a two-stage fine-tuning process:

  • Supervised Fine-Tuning (SFT): Trained on 270 real admissions essays from public collections (JHU "Essays That Worked", College Essay Guy, AP Study Notes) and approximately 19.4k human essays from the open persuade corpus. This stage used LoRA with r=16, a learning rate of 2e-4, and 1 epoch.
  • Direct Preference Optimization (DPO): Utilized the HumanLLMs method (arXiv 2501.05032) where real human essays were marked as 'chosen' and SFT model outputs as 'rejected', alongside GradGPT quality pairs. This stage also ran for 1 epoch with a learning rate of 5e-5.

Key Capabilities

  • Generates authentic college application essays (Common App personal statement style).
  • Produces content with a natural human voice, varied sentence rhythm, and honest reflection.
  • Capable of incorporating specific personal details based on prompts.

Usage Considerations

While trained on human essays for natural style, the model's output is not guaranteed to bypass AI detectors, as these are trained classifiers with variable results. A 4-bit MLX build for Apple Silicon is available at alphanozcan/essAi-9b-mlx.

Finetunes
2 models