notes for software engineers getting up to speed on new AI developments. Serves as datastore for https://latent.space writing, and product brainstorming, but has cleaned up canonical references under the /Resources folder.

《多模态大模型：新一代人工智能技术范式》作者：刘阳，林倞

Use Llama3.2 Vision for tagging and searching images on your local machine.

'Talk to your slide deck' (Multimodal RAG) using foundation models (FMs) hosted on Amazon Bedrock and Amazon SageMaker

AI-powered disaster response platform with offline-first architecture using Gemma 3n. Provides computer vision hazard detection, voice analysis with emergency keywords, PDF report generation, and multi-user coordination - all working without internet access.

Multimodal and multilingual topic model with pretrained embeddings

Supplementary material for "Deep Multi-modal Object Detection and Semantic Segmentation for Autonomous Driving: Datasets, Methods, and Challenges" IEEE transactions on intelligent transportation systems: https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=9000872 -- arxiv: https://arxiv.org/pdf/1902.07830.pdf

Circuitry.ai is an open-source tool that combines computer vision and large language models to detect, analyze, and explain electronic circuit diagrams. It leverages YOLOv8 for component detection and LLaMA 3 for generating intelligent textual explanations of how the circuit works.

Multi Modal Agent using Deepgram and Groq LPU's and Sentence Transformers for Vector Embeddings, ChromaDB for persistent vector db storage and BM25 with ColBERT reranking for hybrid search for RAG context management

Praat++ — A web-based multimodal annotation tool for acoustic and video data.

A simple but cool visualization project of multimodal Emotions-KG based on d3

Code to recreate results from the "Multi-view biclustering via non-negative matrix tri-factorisation" paper.

Medibot is a voice-enabled medical AI assistant using RAG for accurate healthcare conversations. Evolved from my text-based chatbot, it now understands spoken questions and responds with voice answers, making medical guidance more accessible through intuitive multimodal interaction.

Notebooks and scripts used for the Nautilex Hackathon

Real time local video chat example with llama.cpp

Vidi: AI Video Agent for Blind and Low Vision Users

Detect miscontexualization to classify image-caption pairs as Out-of-Context or Not

Framework for training and evaluating "Single-Branch Network Architectures to Close the Modality Gap in Multimodal Recommendation" (TORS Journal Paper)

A Flask-based web app for managing multimodal datasets text and images with CRUD operations via SQLite, and seamless export as a structured Parquet dataset to Hugging Face Hub.