alphanozcan/essAi

Model Overview

essAi is a 4 billion parameter model, fine-tuned from Qwen3-4B by alphanozcan, specializing in generating authentic college application essays. Its primary goal is to produce Common App style personal statements that mimic a natural human voice, incorporating specific personal details, varied sentence rhythm, and honest reflection.

Key Capabilities

• Authentic Essay Generation: Produces college application essays (e.g., Common App personal statements) with a human-like narrative style.
• Natural Language Output: Trained to achieve a natural writing flow, aiming for human-like burstiness and sentence length characteristics.
• Specialized Fine-tuning: Underwent a two-stage fine-tuning process (SFT and DPO) on a dataset of approximately 19.7k human-written essays, including real admissions essays and the open persuade corpus.

Training Details

The model was fine-tuned using LoRA (r=16) for the Supervised Fine-Tuning (SFT) stage and a DPO stage utilizing the HumanLLMs method with real human essays as 'chosen' and SFT model outputs as 'rejected'.

Good For

• College Application Essay Drafting: Ideal for students or educators seeking assistance in generating initial drafts or exploring narrative angles for personal statements.
• Creative Writing Assistance: Useful for tasks requiring the generation of reflective, personal narratives with a specific stylistic target.

Limitations

As a 4B parameter model with one training epoch, its coherence quality reflects these constraints. While trained on human essays for natural style, its output's detectability by AI detectors is not guaranteed, as detector performance varies.