[1] RLHF-Blender

An interactive framework for learning reward models from diverse human feedback, combining pairwise preferences with richer feedback types in one configurable pipeline.

• Built to support research on multi-type human feedback for RLHF.
• Includes an interactive UI for feedback collection and reward model experimentation.

[Repo] [Website] [Demo]

[2] smartgpt-ui

A lightweight web interface for SmartGPT-style prompting workflows to improve reasoning quality with staged prompting and response refinement.

• Inspired by AI Explained's SmartGPT workflow and built on top of Chatbot-UI.
• Provides a practical interface to experiment with multi-step reasoning prompts.

[Repo] [Live Demo]

[3] Tackling the ARC Challenge with Object-Centric World Models

A research prototype exploring object-centric world models and reinforcement learning for François Chollet's ARC challenge.

• Focuses on structured representations for better compositional generalization.

[4] Simple RAG for Confluence

A retrieval-augmented generation workflow to ingest heterogeneous files and produce structured summaries for Confluence pages.

• Designed for practical team knowledge capture and fast internal search.