deepseek-v3

A voice-enabled AI agent built with Google Gemini (Flash) and LangGraph, integrating Speech-to-Text and Text-to-Speech for natural, LLM-driven interactions, delivered through a Streamlit interface. It is capable of scheduling event in google calender.

Godeye AI is an open-source AI system powered by Deepseek, designed to identify emerging topics and correlates them with existing and newly created cryptocurrencies and memecoins, providing valuable insights into market dynamics.

0ns zero-copy, branchless MoE routing infrastructure for Mixtral/DeepSeek-V3. Bypasses NCCL All-to-All collective stalls via XLA shard_map virtual memory address MUX, locking VRAM complexity to static O(1)

Building a Personal Tutor with Spring AI and DeepSeek

Modular prompts and software for calibrating AI reviewer of scientific papers (6 modules: mode detection, novelty, methodology, practical value, visualization, reproducibility)

Examiq is an AI-powered proctoring system that monitors students taking online exams through gaze, input, and behavior tracking.

🛠 Fix DeepSeek-V3 Windows setup issues by resolving CUDA memory errors, missing DLLs, and PyTorch-CUDA conflicts for smooth local deployment.

A Bare-Metal PIM-HBM Hardware Co-Design Core Engine exploring Register-Level MUX Virtualization to mitigate NCCL All-to-All communication stalls.

Linux port of DeepSeek Harness Desktop (AppImage, .deb, Snap) with native UI integration.

A ComfyUI extension for DeepSeek(API)

📰 Automate news research and curation with GraphNews, a multi-agent system that ensures high-quality, factual content through self-correction.

Autonomous coding agent right in your IDE, capable of creating/editing files, executing commands, using the browser, and more with your permission every step of the way.

🎤 Create realistic text-to-speech outputs with advanced voice cloning and design using Alibaba's Qwen3-TTS model for ComfyUI.

Aether（以太）象征着后端与前端之间无缝的数据流。初了解前端 AI Agent。消除“黑盒感”

This project is a web-based PDF question-answering chatbot powered by Deepseek v3 and R1's Language Learning Models (LLMs). Users can upload PDFs, ask questions related to the uploaded documents, and receive accurate responses. Use the link below to try it out!

A full-stack Yelp analytics platform combining a React + TypeScript dashboard with a FastAPI backend to explore business, review, user, check-in, and rating data through interactive charts, CSV-driven insights, and an AI-powered natural-language-to-SQL research assistant (DeepSeek + PostgreSQL).

End-to-end PDF RAG chatbot designed to process complex 100+ page documents. Features intelligent LangGraph query routing, hybrid ChromaDB search (BM25 + HuggingFace), and cross-encoder re-ranking to boost retrieval accuracy. Deployed via Streamlit and FastAPI with full LangSmith trace and latency monitoring.

A simple experiment applying compressible flow principles to soften distributed gradient communication stalls.

Minimal PyTorch re-implementation of DeepSeek V3 with Multi-head Latent Attention, Rotary Positional Encoding (YARN), Mixture of Experts, and KV cache compression.

🔍 Truth Lens: An AI-powered Chrome extension for instant fact-checking. Built with DeepSeek V3, Tavily Search, and Plasmo.