Browse applications built on Anthropic's technologies. Explore PoC and MVP applications created by our community and discover innovative use cases for Anthropic's technologies.
GEOCLUSTER 2.0 is an AI-powered desktop app for geospatial analysis. It retrieves Sentinel-2 imagery, performs GIS image processing, and lets users interact with an AI assistant using natural language for remote sensing workflows.
The measurement instrument for a token-vs-accuracy cascade router.
This is an AI Powered Pipeline that takes your video, analyzes it, divides up into distinct scenes, and then generates caption for each scene in 4 differrent styles.
Two-stage video captioning agent: Chain-of-Verification grounding feeds four distinct-tone captions, with a self-judge regeneration pass and a fail-safe pipeline that guarantees valid, complete output on every run, even under model or infra failure.
Bridge is an autonomous agent that ports CUDA code to AMD ROCm/HIP until it builds and passes its tests, with a mechanical security gate over every AI-written diff. Proven on AMD hardware (Radeon gfx1100), driven by an open-weights model on Fireworks.
Reverse Resume: AI-powered resume analysis scored against YOUR dream job. Upload your resume, get a Gemini-powered ATS score, gap analysis & a personalised 4-week roadmap to close skill gaps — with resilient multi-model AI fallback so it never fails.
An intelligent multi-agent pipeline that uses vision-language models and a novel memory engine to automatically generate temporally-coherent 3D spatial audio for video.
Zero-token-first routing agent: two small local LLMs run inside the container, wrapped in deterministic verification (math executed, code behaviorally cross-checked, logic solver-proven). Only the weakest answers escalate, under a hard 900-token budget.
A CLI tool that detects CUDA-specific Python code that breaks or degrades on AMD ROCm, flagging each issue by severity with a sourced explanation and fix guidance - filling the gap HIPIFY leaves at the application layer.
An aesthetic cognitive captioning workspace and real-time audio telemetry platform powered by Fireworks AI and AMD Cloud computing. Featuring multi-style subtitle tuning, instant Firestore sync, and dynamic interactive Recharts sentiment tracking.
A batch AI agent that classifies each task into 1 of 8 capability categories and routes it to the cheapest Fireworks model tier that can still answer correctly — minimizing total tokens without failing the accuracy gate.
On-prem, citation-grounded contract compliance AI — every answer traces to a real document and clause, or it refuses to answer. Runs Qwen3-14B via vLLM on real AMD ROCm hardware, so sensitive contracts never leave your infrastructure.
iSquare is a multi-model video captioning agent that generates accurate captions in formal, sarcastic, humorous tech, and everyday humor styles using independent analysis, verification, fact-grounded writing, and internal validation.
Voxniac One is a sovereign B2B sales agent for technical founders who hate cold calling. It interviews you, builds your sales profile, and makes live calls—running open-weight LLMs on AMD Instinct MI300X via Fireworks AI, with a Twilio human handoff.
An open-source prosthetic hand that sees what you're reaching for and pre-shapes the grip before contact. The only thing you decide is when to close, not how.
A hybrid routing agent that answers easy tasks with a bundled local model at zero token cost and escalates only accuracy-critical categories to Fireworks frontier models — cutting API spend while clearing the accuracy gate.
An adaptive and highly efficient hybrid LLM routing agent designed for the AMD track. It dynamically optimizes inference costs and latency by routing diverse operational tasks to the most suitable open-source models while keeping a zero-failure rate.
Built three AI hackathon agents: a cost-efficient task router (local-first, escalates to Fireworks only when needed), ROCm Ready (ports CUDA to AMD GPUs with hipify), and a local video captioning agent generating four caption styles per clip.
Plainlode reads live market demand and supply for any product category and returns one plain-language decision on what to sell next, not a dashboard. Built on Fireworks AI and AMD Instinct.
Kernel Olympics is a multi-agent AI pipeline that automatically ports NVIDIA CUDA kernels to AMD ROCm/HIP — transforming weeks of manual engineering into ~2 minutes per kernel with 3-stage validation gates and real GPU verification.
testtesteststesttesteststesttesteststesttesteststesttesteststesttesteststesttestests
An intelligent AI gateway that uses LangGraph to route requests across local Gemma, GPT, Claude, and Gemini models based on cost, latency, privacy, quality, and token efficiency, with explainable routing and analytics.
Building Human-like AI companions with human-inspired cognition and self-learning capabilities that grow, adapt, and evolve through experience.
Shadow QA is an autonomous testing agent that uses Vision-Language Models to navigate web apps, visually identify UI/UX bugs, and generate comprehensive QA reports without hardcoded scripts.
The only video captioning studio that QCs itself. Gemma locks the facts, then writes four calibrated voices — formal, sarcastic, tech & non-tech humor — each caption pre-judged for accuracy and tone before you see it. Gemma-first on Fireworks AI.
swarmtrace is a ai agent monitring platform build for to give you complete control of your ai agents saas
TokenEdge is a verification-gated AI routing agent that solves tasks locally whenever possible and uses Fireworks AI only when necessary, minimizing scored tokens while maintaining accuracy.
RocPorter automates CUDA-to-ROCm migration by converting CUDA to HIP, analyzing compatibility, estimating engineering effort, and producing detailed migration reports that help developers confidently move workloads to AMD GPUs.
A containerized AI agent that routes tasks across 8 capability categories to the cheapest sufficient Fireworks AI model, using zero-token local classification and syntax validation to minimize token spend while clearing the accuracy gate.