Anthropic

Top Builders

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

Anthropic

Anthropic’s Constitutional AI training approach research focuses on developing AI systems safe by design and aligned with human values. By prioritizing safety, we can create strong and corrigible AI systems that are safe for humans to use.

General
Company	Anthropic
Founded	2021
Discord	https://discord.gg/lablab-ai-877056448956346408

Anthropic Claude

Claude is your friendly and versatile AI language model that can assist you as a company representative, research assistant, creative partner, or task automator.

Claude is Safe, Clever, and Yours. Built with safety at its core and with industry leading security practices, Claude can be customized to handle complex multi-step instructions and help you achieve your tasks.

You can easily use Claude for your app, and all necessary APIs, boilerplates, tutorials explaining how to do so and more, you can find on our Claude tech page.

Claude Code

Claude Code is a command-line tool from Anthropic for agentic coding. It enables Claude to refactor, debug, and manage code directly in the terminal. You can find more information on our Claude Code tech page.

Edit on GitHub

Anthropic AI Technologies Hackathon projects

Discover innovative solutions crafted with Anthropic AI Technologies, developed by our community members during our engaging hackathons.

WritenDraw Flight Simulator for Software Developer

WritenDraw is an agentic AI simulation platform that puts junior developers through realistic production incidents to bridge the gap between learning to code and working in a real team. The core innovation is the agentic workflow: Google Gemini 2.0 Flash orchestrates the entire simulation through three autonomous agents: AGENTIC EVALUATION: Every step requires free-text responses (no multiple choice). Gemini evaluates each response against per-step rubrics, scoring reasoning 0-15. The AI adapts feedback based on accumulated performance. AGENTIC MENTORING: The AI mentor maintains persistent context, tracking understanding level, chat count, and time pressure. Early messages: patient, asks "what do you think?" By message 7+: "just write it up." The agent autonomously decides how much help to give. AGENTIC AUDIT: The system logs every response, chat message, code submission, and score — creating a complete picture of how a developer thinks through a crisis. The AI continuously assesses and adapts. The simulation drops you into a P1 incident at ShopRight (fictional UK supermarket). You join a standup, read a Jira ticket, investigate messy code with no hints, chat with the AI mentor, write a fix, respond to code review, create a deployment plan, and contribute to a retro. Paste is disabled — Key insight: explanation scores higher than code (10 vs 5 points). Wrong code with a great explanation beats perfect code with no explanation — because in real teams, communication matters as much as code. Built on the author's published research — "TrueSkills: AI-Resistant Assessment Through Personalized Understanding Validation" (SSRN, 2025, DOI: 10.2139/ssrn.5674130) — which demonstrated that AI-resistant assessment requires evaluating understanding rather than recall. WritenDraw takes this further: testing how developers think under realistic production pressure. Built with Python/Flask, Google Gemini 2.0 Flash, CodeMirror, Pyodide, and Docker.

OpsTwin AI

OpsTwin AI is a simulation-first autonomous warehouse control system designed to model and optimize multi-robot fulfillment operations. As warehouses adopt robotics at scale, fleet coordination becomes increasingly complex. Congestion, battery constraints, task prioritization, and workload balancing impact throughput and efficiency. OpsTwin AI addresses this by creating a digital twin of warehouse operations, a live simulation where robotic workflows can be orchestrated, tested, and optimized before real-world deployment. In OpsTwin AI, robots operate within a simulated warehouse grid containing storage racks, charging stations, and pack zones. When a new order is created, the system autonomously determines which robot should fulfill it. Instead of relying on hardcoded rules, I use Gemini as a strategic planning layer. The backend sends live fleet state, including robot positions, battery levels, and active tasks, to Gemini. Gemini returns structured JSON with a selected robot and step-by-step task sequence. This allows deterministic execution while enabling adaptive multi factor decision making. The Vultr-hosted backend serves as the centralized system of record. It maintains robot state, order queues, and operational metrics, and broadcasts real-time updates to a web dashboard using WebSockets. A 500 millisecond simulation loop executes plans, updates robot movement, tracks congestion events, and manages battery-aware charging. The result is fully autonomous multi-robot operation without manual intervention. From a business perspective, OpsTwin AI functions as an operational control tower for robotic fleets, enabling teams to simulate workflows, evaluate performance, and reduce deployment risk before scaling to physical infrastructure. By separating AI planning from deterministic execution, the architecture mirrors real world robotics systems and provides a clear path from simulation to real-world deployment.

DroneOS - AI-Powered Autonomous Fleet Dispatch

DroneOS is an autonomous drone control framework built on PX4 Autopilot and ROS2. At its core is drone_core, a custom C++ SDK that exposes high-level flight control as ROS2 services — arm, takeoff, position commands, land. An OpenClaw AI agent runs on a Vultr VPS and acts as the fleet dispatcher. When emergency incidents come in through the dispatch service, they're routed to the agent via a bridge over WebSocket. The agent evaluates incident priority, checks drone availability and location, then sends flight commands through ROS2 to dispatch drones autonomously. The architecture is two servers connected over Tailscale VPN. The Vultr VPS runs the OpenClaw gateway, dispatch service, communication bridge, and React frontend. A separate simulation server runs PX4 SITL with Gazebo, dual drone_core nodes, rosbridge, and camera feeds. This is the same split you'd have in production — cloud command center talking to drones over VPN, except the drones are simulated. The frontend is a real-time dashboard connected to rosbridge over WebSocket. It shows the incident queue with priority levels, a map with drone positions, live camera feeds from both drones with picture-in-picture toggle, and an AI activity log showing every decision the agent makes. Operators see what the AI is doing and can override with natural language commands through the same OpenClaw agent. The dispatch service simulates a 911 CAD system generating incidents — medical emergencies, fires, property damage — each with priority levels and coordinates. The AI doesn't follow scripts. It decides which drone to send based on priority, proximity, and availability. The framework supports real hardware. Production Docker configs exist for Raspberry Pi companion computers communicating with Pixhawk flight controllers over serial. The simulation runs the same software stack. Live demo: http://207.148.9.142:3000 Source: https://github.com/ortegarod/drone-os

SociClaw

The Problem: Creating high-quality social media content consistently is exhausting. Most creators and brands struggle with the daily loop of ideation, writing, and asset generation. Existing tools are just schedulers; they don't do the work for you. The Solution: SociClaw SociClaw is an autonomous AI agent running on OpenClaw that acts as your dedicated Social Media Manager. It doesn't just post; it thinks. Core Workflow: 1. Research: The agent scans trending topics and news relevant to your niche. 2. Planning: It builds a content calendar (e.g., "Educational Thread on Tuesday", "Meme on Friday"). 3. Generation: It drafts hooks, bodies, and CTAs, and generates custom visuals using AI image models. 4. Sync: Instead of posting blindly, it syncs drafts to your Trello board or Notion database for human review (Human-in-the-Loop). Technical Stack: • Framework: OpenClaw (Agent Runtime) • Integrations: X API, Trello, Notion • AI Models: Large Language Models for copy, Diffusion models for images • Blockchain: Exploring Solana/Base for high-speed, low-cost on-chain content verification and agent-to-agent payments. SociClaw turns "posting every day" from a chore into a fully automated background process, allowing creators to focus on strategy while the agent handles execution.

RIFFAI

RIFFAI’s hackathon prototype demonstrates an end-to-end pipeline for energy site intelligence. The prototype ingests multi-source satellite imagery (Google Earth Engine + direct providers), weather feeds and customer training sets; applies an ultra-fast image/geo-metadata alignment pipeline; runs geospatial segmentation (U-Net/CNN) and time-series forecasting (LSTM); and produces explainable suitability scores and a ranked list of candidate polygons for renewable energy deployment. A minimal React web UI or API visualizes results and emits a sample monitoring alert (thermal anomaly or vegetation/water change). The goal is to show that fast metadata processing + energy-tuned models produce practical, developer-ready outputs in near-real time. We’ll provide domain guidance, sample tiles and a small labeled set for teams that request it.

AI-Powered Transaction Monitoring

SAR Copilot is an AI-powered transaction monitoring and investigation platform designed to solve one of the most persistent challenges in financial crime compliance: overwhelming alert volumes combined with low detection quality. Traditional transaction monitoring systems rely heavily on static, rule-based logic. These systems generate thousands of alerts every week, yet the majority are false positives. Analysts spend most of their time reviewing benign activity while sophisticated fraud patterns remain hidden in the background. SAR Copilot reimagines this workflow by using AI to reduce noise, surface high-confidence cases, and provide full investigation context with explainable reasoning and automated regulatory reporting. At its core, SAR Copilot is built to answer a critical question: how can financial institutions move from reactive, rule-driven alerting to proactive, intelligence-driven fraud detection that works in real time and scales with complexity. Instead of flagging individual transactions in isolation, the system evaluates behavior, relationships, timing, and context across accounts and networks. The result is not a list of alerts but a curated set of investigation-ready cases. In conclusion, SAR Copilot is not just an alerting tool. It is a full transaction monitoring and investigation platform built for modern financial crime challenges. By integrating behavioral analytics, graph-based network detection, unsupervised discovery, contextual scoring, real-time intervention, and automated SAR generation, the system delivers measurable improvements in detection accuracy, analyst efficiency, and regulatory readiness. From raw transaction data to regulator-ready reports, SAR Copilot shows how AI can transform compliance workflows and significantly reduce the gap between detection and action.