embedding-model

LSTM with attention for time series predictions of stock prices using own Ticker Embedding model

おかえり (welcome back): a calm homepage for your favourite science & tech feeds, ranked on your own machine.

langchain-chat is an AI-driven Q&A system that leverages OpenAI's GPT-4 model and FAISS for efficient document indexing. It loads and splits documents from websites or PDFs, remembers conversations, and provides accurate, context-aware answers based on the indexed data. Easy to set up and extend.

In-process C++ embedding generation using ONNX and tokenizers

PyTorch Implementation of the paper "Semantic Compression Embedding for Generative Zero-Shot Learning" accepted to IJCAI-2022

App that answers questions about any YouTube video using its transcript. Built using Retrieval-Augmented Generation (RAG) with LangChain, vector stores, and LLMs.

An end-to-end model for variable-size discovery and representation of atomic propositions (EMNLP 2026)

Turning audio embeddings into speech recognition: a frozen EmbeddingGemma 2 audio encoder plus a 9.66M-parameter GRU-CTC reader, no encoder fine-tuning

This is a research project for movie recommendation system using RL

CRAG -A pipeline that uses tunable thresholds to validate document relevance, refines content at sentence level, and generates citation‑aware answers exclusively from verified sources avoiding hallucinations.

A lightweight AI-powered YouTube assistant built with FastAPI, FAISS, LangChain, and Google Gemini. Extract transcripts, build vector embeddings, and ask natural language questions about any YouTube video — all running locally through a Chrome Extension frontend.

SCYTH-J is a neural network-based web application for classifying Japanese texts into user-defined dynamic topics. The system uses a quantized supervised trained deep learning model and natural language processing (NLP) techniques to analyze and categorize texts.

PHDM 21-Dimensional Embedding Model — Poincaré Hyperbolic Disk with 6-tongue position, 6-phase, 9-telemetry canonical state.

It relates with implementation of RAG pipelines. Quest to fine tune different elements(retrieval , generation broadly) of a domain specific RAG and analyze change from base results. Further experiments on RAG frameworks, prompt enineering and more.

A UTCP manual for the Google Gemini embedding model, allowing any UTCP-compliant client to generate text embeddings.

Local AI Semantic Search

Isomorphic ultra edge AI embedding model + inference library using ONNX Runtime for vector embedding with exact and TurboQuant-style vector search in-browser, Node.js on CPU + GPU in < 40 MB model with R@5 retrieval in < 100ms.

This project implements a full digital thread intelligence system. It supports natural-language engineering queries, part replacement suggestions, compatibility evaluation, and intelligent search across manufacturing assemblies.

NITW Chatbot is a Retrieval-Augmented Generation (RAG) based AI system that answers queries using official institutional documents. It scrapes PDFs, generates embeddings, stores them in a FAISS vector index, and retrieves relevant context for LLM-based response generation, ensuring grounded and accurate answers.

This project is a Spring Boot chatbot that lets users upload their own files and ask questions about them, using Retrieval-Augmented Generation (RAG) with a local Ollama LLM.