Top Builders

Explore the top contributors showcasing the highest number of app submissions within our community.

DeepSeek V3

General
Release date2024
AuthorDeepSeek
WebsiteDeepSeek Models
Repositoryhttps://github.com/deepseek-ai
TypeMoE (Mixture of Experts) Language Models

The DeepSeek V3 model represents our most advanced AI architecture, designed for complex reasoning tasks and code generation. With enhanced context handling and improved instruction following, this model excels in technical applications and enterprise deployments.

Key Features

  • DeepSeek-V3: 671B parameters (37B activated per token), optimized for math, code, and multilingual tasks.
  • Code Generation: Supports 12+ programming languages
  • Advanced Reasoning: Chain-of-thought capabilities for multi-step problems
  • Enterprise-Grade Security: Built-in content filtering and compliance features
  • Speed: 3x faster generation than previous versions (60 TPS)
  • Open-Source: FP8/BF16 weights available on Hugging Face

👉 Local Deployment Guide for DeepSeek V3 👉 Model Weights on Hugging Face 👉 API Documentation 👉 Deepseek V3 Paper 👉 Performance Highlights

Edit on GitHub

Deepseek DeepSeek V3 AI technology Hackathon projects

Discover innovative solutions crafted with Deepseek DeepSeek V3 AI technology, developed by our community members during our engaging hackathons.

AuraOS

AuraOS

Aura is a sovereign, neuro-symbolic AI operating system designed to make advanced artificial intelligence more transparent, efficient, collaborative, and accountable. Instead of sending every instruction directly to a language model, Aura can locally interpret natural-language intent, organize it into structured semantic and routing representations, identify relevant code or knowledge, and prepare a bounded task before model use is required. Aura combines deterministic routing, a 4,096-primitive lexical system, polysynthetic six-slot intent structures, visual code topology, guarded human-agent workflows, and a Learning Arena where verified experiences can be tested before becoming reusable knowledge. Its Observatory allows users to see how Aura understood an instruction, while the Human Agent Arena supports node-specific dialogue, approval gates, verification, and preserved attempt artifacts—including failed work that may help future refactoring. Aura is built around human authority. Models can propose, explain, investigate, and assist, but they cannot silently expand scope, bypass evidence requirements, or merge changes without approval. The result is an AI architecture focused not only on capability, but on sovereignty, inspectability, learning, and responsible action.

ResearchOS

ResearchOS

ResearchOS ResearchOS is an autonomous multi-agent research platform designed to accelerate academic and technical research. Instead of acting as a traditional chatbot, ResearchOS coordinates a team of specialist AI agents that work together to perform an end-to-end research investigation. Given a research question, the system: -plans the investigation -retrieves relevant academic papers from arXiv -ranks and analyzes the strongest evidence -synthesizes findings across multiple papers -identifies evidence-grounded research gaps -generates a structured research report -stores completed investigations in notebook memory The platform presents the entire workflow through an interactive interface that visualizes each stage of the investigation, making the research process transparent and easy to follow. Key features include: -Multi-agent orchestration -arXiv evidence retrieval -Cross-paper literature synthesis -AI-generated research reports -Automated research gap discovery -Persistent notebook memory -Interactive investigation timeline -Deterministic fallbacks for reliable execution Although ResearchOS supports any research domain, it was developed with AMD-focused technical research in mind and can investigate topics such as ROCm, MI300X, GPU optimization, and LLM inference.

Aora - The Adaptive, Optimized Routing Agent

Aora - The Adaptive, Optimized Routing Agent

Here's a version that's about **1,700 characters**, giving you ~300 characters of buffer under a 2,000-character limit. --- ## **Problem** Modern AI applications struggle to choose the right LLM for each request. Static or manual model selection increases cost, latency, and maintenance while producing inconsistent results as models rapidly evolve. ## **Solution** **AORA (Adaptive & Optimized Routing Agent)** is a natively AI-orchestrated routing engine for Fireworks AI. It profiles each request and dynamically selects the optimal model to maximize quality while minimizing cost and latency. ## **How It Works** Built in Python, AORA integrates with the Fireworks AI inference engine. An AI-augmented Task Profiler identifies the request type (reasoning, coding, summarization, translation, extraction, etc.). The Orchestration Engine scores eligible models using configurable utility metrics such as capability, latency, and cost, then routes the request automatically. Configuration is environment-driven with optional runtime model discovery. AORA is Docker-native, supports offline image builds, and is deployable locally, in the cloud, or via Docker Hub. ## **Key Features** * AI-augmented task profiling * Adaptive utility-based routing * Native Fireworks AI integration * Runtime model discovery * Real-time telemetry * Docker-native deployment * API-first architecture ## **Market Opportunity** AORA targets AI developers, startups, SaaS platforms, and enterprises building multi-model AI applications. As organizations adopt multiple foundation models, intelligent routing becomes critical infrastructure for optimizing quality, speed, and cost. ## **Roadmap** Future releases will add reinforcement learning, semantic caching, budget-aware routing, A/B model evaluation, multi-provider support, enterprise governance, and autonomous optimization, evolving AORA into an adaptive control plane for multi-model AI systems.

Agents Terminal V3

Agents Terminal V3

AGENTS TERMINAL V3 — Self-Improving AI Agent Orchestration Every interaction makes your AI team smarter. Every business on WhatsApp faces the same problem: customers ask the same questions daily: hours, pricing, orders. The chatbot answers some, fails on others, and the gaps repeat tomorrow. It never learns. For African SMEs this is critical, load-shedding kills uptime, no engineering team exists, and WhatsApp is the primary business channel running on a static AI. Agents Terminal V3 implements a compound loop: every customer message flows through n8n (84+ nodes, 6 workflows) → An LLM 7-intent classifier (200ms) → dual capture to Redis Cloud + PostgreSQL → daily artifacts across 4 time windows → 5-step memory loop (episode → lessons → promote → revise → index) → weekly gap analysis → weekly upgrade pipeline (A 7 phases pipeline generating FAQ suggestions, knowledge base entries, prompt refinements, handoff reviews) → human reviews → applies → smarter customer service the following week. Architecture: Docker containers for n8n + PostgreSQL. Redis Cloud for real-time streams. DeepSeek v4-flash for classification. Twilio + Cloudflare for WhatsApp ingress. 16-command orchestrator CLI. 4 JavaScript tools (capture.js, 3lm.js, review.js, promote.js). 3-Layer Memory: episodic, semantic, procedural — file-first, markdown-first. 19 connections. 5 subsystems. 0 broken links. The loop compounds every week. Built with AMD GPU Cloud credits + Fireworks AI for AMD-optimized inference path. Team: AI-Suplex Labs GitHub: github.com/kmagwenzi/Agents-Terminal-V3

Vektor

Vektor

Autonomous artificial intelligence agents are increasingly deployed to automate critical business operations, yet their probabilistic nature introduces significant operational risk. When an agent interprets ambiguous instructions, organisations face potential financial and regulatory disasters. Vektor solves this problem by acting as a deterministic Guardian Runtime that sits outside the process boundary of the primary agent. Instead of relying on fragile prompt engineering or output guardrails, Vektor physically intercepts tool calls before any real-world side effect can occur. The core design invariant of Vektor is that the safety decision path contains no language models. While the planning agent interprets unstructured text to propose actions, the runtime engine evaluates those proposals against declarative enterprise policies using only table-driven comparisons and verifiable evidence lookups. If an action violates a policy or lacks required compliance credentials, Vektor immediately blocks execution and appends the decision to an auditable hash-chained log. When possible, the system automatically feeds the failure requirements back to the agent so it can remediate the action and execute safely. Because the core evaluation engine is completely domain-agnostic, organisations can scale this exact security infrastructure across procurement workflows and regulated financial systems simply by swapping the underlying policy packs. Engineered for high-performance deployments on AMD Instinct hardware using ROCm acceleration, Vektor ensures sub-second inline verification without bottlenecking enterprise operations. Ultimately, it transforms artificial intelligence safety from a probabilistic guessing game into reliable and verifiable infrastructure.

VigiAI focused on the alert feature

VigiAI focused on the alert feature

VigiAI is an intelligent real-time surveillance system for detecting traffic accidents and anomalies. Running on an AMD MI300X accelerator, its hybrid architecture splits tasks between local processing—using Qwen-VL 7B, YOLO v8x, optical flow, and pose estimation—and cloud analytics via DeepSeek-V4 Flash through the Fireworks AI API. To handle chaotic real-world events, it was trained on the complete UCF-Crimes dataset (over 100 GB of footage). The system processes video through a structured 7-phase pipeline: Event Detection: Identifies anomalies instantly. Localization: Pins down spatial bounding boxes and timestamps. Scene Analysis: Evaluates weather, road, and lighting conditions. Causal Inference: Infers accident causes and secondary risks. Object Tracking: Tracks and counts vehicles or pedestrians. Speed Estimation: Measures velocity using pixel-to-meter calibration. Narrative Generation: Compiles an executive summary and detailed report. VigiAI detects 14 anomaly classes (accidents, theft, fights, fires, etc.) and transmits instant alerts via WebSockets over a public Cloudflare Tunnel. The web dashboard features live camera/webcam feeds, a GPS map, historical logs, and local vs. cloud performance metrics.