TOBIAS - better web pages through intelligent routing and judgement
Alexander Pinchuk · Team TOBIAS
Submitted to Apart x Martian Mechanistic Router Interpretability Hackathon. Sprint projects are early-stage work by participants, not Apart Research publications.
We implemented routing and judging system designed to generate and evaluate improved versions of web page content for SEO, advertising and generative search goals. Our system selects the most promising content version using real-world performance indicators, providing a reproducible method for LLM benchmarking and semantic page optimization. This work contributes to Track 1 (Judge Model Development) of Apart x Martian Mechanistic Router Interpretability Hackathon, while integrating Judge Model insights for end-to-end performance routing.
Reviews
Strong domain-specific project. Curious how it would generalize to other multi-objective generation tasks. I'd add at least one MI probe to the implementation to see how it guides generation.
Constructive feedback:
Strength:
Nice application:web pages optimisation via content extraction
Love the data processing pipeline
Uses a post hoc routing approach
Nice judges with an application focus to website/ads.
Weakness:
Comparison against a “raw prompting” baseline like feeding website content and an instruction to the models would be good to justify this routing pipeline.
Expert Orchestration: 4
MI: 1
Tech Imp and rep: 4
- Seems useful if you want a good website builder for a business
- Actual safety impact seems negligible
- Seems fine technically
Cite this project
@misc{pinchuk2025tobias,
title = {{TOBIAS - better web pages through intelligent routing and judgement}},
author = {Alexander Pinchuk},
year = {2025},
month = jun,
note = {Submitted to Apart x Martian Mechanistic Router Interpretability Hackathon, an Apart Research Sprint},
howpublished = {\url{https://apartresearch.com/sprints/projects/tobias-better-web-pages-through-intelligent-routing-and-judgement-ea8h}},
url = {https://apartresearch.com/sprints/projects/tobias-better-web-pages-through-intelligent-routing-and-judgement-ea8h}
}More from Apart x Martian Mechanistic Router Interpretability Hackathon
- 1st place by peer reviewView project: Manipulating Self-Preference for Large Language Models
Manipulating Self-Preference for Large Language Models
Team Preference
Large language models (LLMs) carry great value as evaluators of synthetic data for research and production settings. However, recent research shows that language models exhibit bias towards their own responses in blind …
- 2nd place by peer reviewView project: Approximating Human Preferences Using a Multi-Judge Learned System
Approximating Human Preferences Using a Multi-Judge Learned System
AutoBox
In this work, we introduced a learned approach to aggregating multi-judge scores: using a GAM and a simple MLP as an alternative to traditional, non-learned methods like averaging. Our models outperform the naive …
- 3rd place by peer reviewView project: Judge using SAE Features
Judge using SAE Features
SAEwhat?
The key idea of this project was to explore model judgement using Sparse Autoencoder (SAE) features for mathematical reasoning tasks involving addition, multiplication, and subtraction operations. We compared this …