Project
Something built and shipped.
Releasing a new sentiment classification model.
Our new sentiment model is free on Hugging Face. It scores every heading, list item and sentence from very positive to very negative, gets 84% of labels exactly right and 95% within one point, and is built into DEJAN App and a public demo.
Introducing LinkjeBERT a Dutch Language Model
LinkjeBERT is a Dutch language model trained on structured Markdown to predict where humans will place links in text using token-level confidence levels.
Teaching a Model to Reason Before It Learns to Talk
An exploration of building tiny, logic-first models using cellular automata to challenge the transformer paradigm and identify the primitives of reasoning.
TurboQuant: From Paper to Triton Kernel in One Session
An implementation and technical analysis of Google's TurboQuant algorithm, testing KV cache compression on Gemma 3 4B using PyTorch and custom Triton kernels.
Reverse Prompting: Reconstructing Prompts from AI-Generated Text
A fine-tuned Gemma 3 270M model reconstructs the most likely prompts from AI-generated responses using synthetic data and contrastive search configurations.
Grounding Snippet Extraction Tool
The Gemini Grounding Tool identifies which URLs and specific sentences Google's AI extracts to ground its answers, helping optimize content for AI search.
Fan-Out Query Search Volume Prediction Using Deep Learning
A deep learning approach using a Query Demand Estimator to automatically predict search volume ranges for long-tail queries generated by a fan-out model.
Introducing Tree Walker
Tree Walker is an analysis tool designed to deconstruct how AI models like Gemini perceive brands by uncovering word uncertainty and probabilistic language paths.
Universal Query Classifier
A zero-shot, multi-label search query classifier that maps queries to any user-provided label taxonomy without the need for retraining or bespoke models.
Multi-Step Research Agent
An implementation of Google's query fan-out in an agentic framework used to research the machine learning and SEO services offered by DEJAN Marketing.
Content Substance Classification
Cyberfluff is a novel approach for detecting low-quality web content using curriculum-driven contrastive pretraining to distinguish fluff from substance.
AI Content Detection
DEJAN-LM is an AI content detection model trained on 20 million sentences, using a combined deep learning and heuristic approach to identify advanced AI text.
Introducing Grounding Classifier
An analysis of Gemini 2.5 Pro's search grounding capabilities and the development of a prompt grounding classifier trained on 10,000 collected prompts.
Introducing VecZip: Embedding Compression Algorithm
VecZip is a novel compression method by DEJAN AI that reduces embedding dimensionality by retaining unique dimensions to improve AI performance and storage.
ILO
The ILO app is a Streamlit-based tool for managing SEO data through URL population, GSC data fetching, query intent classification, and traffic projections.