huggingface Applications

Browse applications built on huggingface's technologies. Explore PoC and MVP applications created by our community and discover innovative use cases for huggingface's technologies.

ROCm Plate & Sign OCR (AMD AI League)

An OCR container for AMD AI League Match 2 that reads licence plates and road signs with an open vision-language model running on an AMD GPU through ROCm. How it works: the container starts a resident server that loads Qwen2.5-VL-7B-Instruct once, in bf16 on the GPU, runs a warm-up inference and then listens on a unix socket. Each call to app.py is a thin client, so the harness can start a new process per image without the model ever being reloaded. Image handling accepts PNG, JPEG and TIFF (including 16-bit and multi-page files), applies EXIF rotation, and resizes very large or very small images to keep the number of visual tokens, and the time per image, bounded. The prompt encodes the transcription rules: return the plate number without state names or slogans, keep the province character and letter on Chinese plates, keep every word printed on a sign, and join multi-line text top to bottom. Post-processing removes labels, quotes and explanations, and folds look-alike separators and full-width characters that the grader's normalization would not remove. Reliability: the model weights are baked into the image, so it runs with no network access. Every call writes a valid JSON file, even when something fails. The build checks that torch and torchvision are ROCm builds. The image is built and contract-checked in GitHub Actions, and evaluated against the sample images plus blurred, noisy, glare, low-light and skewed variants.

Clevermiss

AMD ROCmHuggingFace HubQwen3-Coder

WaltTrace — Offline Multimodal RAG

WaltTrace answers short factual questions from mixed engineering documents while returning the exact files needed to support each answer. It parses PDF, Word, Excel, CSV, logs and source files, and uses a local vision-language model to read diagrams, scanned pages and equipment labels. Retrieval follows identifiers between incident reports and lookup tables, allowing answers that require more than one document. Generated quotations are checked against indexed source text; unsupported answers are refused with an empty answer and citation list. Qwen2.5-VL-7B runs on a real AMD GPU through ROCm PyTorch. All weights are embedded in the mandated container base, and the application runs without outbound network access. Indexing starts a resident model behind a private Unix socket so each fresh question process avoids reloading the weights. Malformed, unreadable, encrypted and unsupported files are skipped independently. The development report documents actual GPU timings, memory readings, individual answers and citation checks. Public-sample results are development evidence, not a claim about hidden evaluation scores or League points already awarded.

WaltTeam

AMD ROCmHuggingFace Hub

GreYat Web Scout: grounded web retrieval

GreYat Web Scout is a Match 4 agent for intelligent web information retrieval. It takes a natural-language question, with an optional start URL, and returns one exact value together with the page it came from. How it works: a language model (Qwen3-4B-Instruct, served on the AMD GPU by a resident process so each query skips the load time) first plans the lookup: search queries, likely official URLs, and package or project names. In parallel the agent queries several search engines and Wikipedia, and for "latest version" questions it reads the authoritative registries directly (PyPI, npm, GitHub releases, endoflife.date). It fetches the best pages, keeps the passages that match the question, and asks the model to extract a single value and say which passage it came from. Grounding comes first. An answer is only returned if the value appears verbatim on the cited page, so the source URL always backs up the answer. If no page confirms it, the agent returns an empty answer instead of a guess, and answers to "latest" questions are never taken from the model's memory. Version strings and numbers are normalised (v-prefixes, LTS labels, thousands separators) to match the grader. Everything runs locally on AMD ROCm with no external API keys. In development, on a 30-question set covering facts, ports, people, dates and live package versions, the pipeline answered 26 correctly (tested with a hosted stand-in model, since GPU time was not available).

GreYat

AMD ROCmQwen3HuggingFace Hub

WebScout Exact: Match 4 Retrieval Agent

WebScout Exact is my entry for Match 4 of the AMD AI League (Intelligent Web Information Retrieval). The agent takes a natural-language question, optionally with a starting URL, and returns one exact value plus the source page it came from, not a summary or a list of links. It runs as a single Docker container built on the official ROCm PyTorch image, so it works on AMD Instinct GPUs. Pipeline: the question is classified into an answer type (number, date, version, price, name, phone and so on); the agent fetches the given URL or searches the web, follows the most relevant links, and cleans each page into plain text. Candidate values are extracted with rules for each answer type and checked by a local Qwen3-4B-Instruct model served from the container on the AMD GPU, which picks the value that best matches the question. The answer is normalised the way the grader compares answers and written to a JSON file with the answer, source URL and confidence. If the model is not available, a rule-based path still answers. Each query is designed to finish well under 30 seconds and the model loads during the startup window. On my own held-out test set it answers 7 of 8 questions correctly. No API keys are used: all reasoning runs locally on AMD hardware. Image: ghcr.io/solar-depin-hub/amd-match4-agent:v1

Vadym Bilobrovets

AMD Developer CloudAMD ROCmQwen3+1

ROBOGUARD DX

RoboGuard DX is an enterprise-grade diagnostic terminal designed for modern industrial automation.

Nexus Team

Gemini 3 FlashGemini AIHuggingFace Spaces

Triage Relay

Not a rule tree and not a chatbot. Two tiny transformers turn a patient's own words into structure; counted, readable "seesaws" route them at the front desk, keep the doubt visible, and hand unsure patients to a person.

VINE

HuggingFace Hub

ReKit — Visual Recovery Intelligence

ReKit connects visual evidence, human review, deterministic compatibility and recovery optimization to traceable kit records. Local NVIDIA inspection and synthetic recovery are demonstrated; AMD execution remains pending.

NeuralFoundry

Qwen3-VLHuggingFace Hub

Community Connect

CommunityConnect helps people in Genova and Liguria find practical, sometimes paid, local support, like tutoring or help with online forms. Seekers describe a need, review AI-suggested details, and see eligible providers ranked with an explainable score.

Community Connect

AIMLAMD Developer CloudAnthropic Claude+8

HealthBridge — A Clearer Path to Care

HealthBridge transforms patient concerns and scattered medical records into source-linked visit briefs that clinicians can review, correct, and turn into clear patient instructions—without automated diagnosis.

Priyanshu Sharan

AMD ROCmHuggingFace SpacesMongoDB+2

quant-agents: Autonomous Quant Research on AMD

Four AI agents on one local Qwen MoE served by vLLM on AMD MI300X brainstorm trading ideas, write vectorized backtests, audit them for lookahead and overfitting, and package survivors for C++ production. No cloud APIs.

quant-agents

AMD Developer CloudAMD ROCmQwen3-Coder+4

RADIANT X-Ray Triage & Evidence Assistant

RADIANT compares chest X-ray model findings against report findings and surfaces disagreements for human review.

Zehra-coded

AMD Developer CloudAMD ROCmLLM Gateway+2

ConsultPrep AI

ConsultPrep AI helps patients prepare for doctor visits by organizing their own descriptions of symptoms, timing, and daily impact into a clear, editable summary they can review and share with a clinician.

del Pilars

VercelAMD Developer CloudChatGPT+5

MediGuide AI Multilingual Healthcare Intelligence

An AI-powered healthcare assistant that extracts, summarizes, and explains medical reports in multiple Indian languages, helping patients understand clinical information and prepare for doctor consultations.

Team Tech

AgentOpsAI/ML APIAMD ROCm+7

AutoShop, Connecting agents with customers-retail

Autonomous Store sells and restocks itself. Gemma 4 on AMD powers a shopper agent that finds, compares, and buys products, while a ROCm-ready demand forecast reorders stock before it runs out, so fewer sales are lost to stockouts.

IJustPressButtons

AI/ML APIAMD ROCmAWS+4

BlackBox — Persistent Visual Memory

BlackBox gives cameras persistent memory, tracking objects across time and reasoning under uncertainty to answer questions about what likely happened when an object was no longer visible.

BlackBox

AMD Developer CloudAMD ROCmHuggingFace Hub+1

CLARIO

CLARIO turns years of scattered symptoms and medical records into a clear story for doctors. Powered by AMD AI, it combines specialized models, rare-disease knowledge, and human oversight to help patients be heard sooner - without replacing clinicians.

Revived

AMD Developer CloudDeepmindGemma+3

TradeForensics: Commerce Action Copilot

AI diagnostics copilot connecting transaction logs, carrier fulfillment latency, and customer reviews into deterministic root causes and actionable operational remediation plans — accelerated by AMD Instinct MI300X.

RyzenShine

AgentOpsAIMLAI/ML API+9

KGB: Grounded OCR and RAG on AMD

KGB provides two GPU inference containers for the AMD AI League: optical character recognition and document-grounded question answering. Both preserve the required AMD ROCm PyTorch base layers and include a pinned Qwen3-VL-8B-Instruct model, so inference does not need a network connection or a hosted API. The OCR service reads license plates and road signs from PNG, JPEG and TIFF inputs. It keeps the model resident while each evaluator invocation writes its own JSON result. The RAG service indexes a mixed corpus once, reads PDF, Word, spreadsheet, CSV, source-code and image content, and returns a short answer with corpus-relative citations. It validates quoted evidence and returns an empty answer when support is missing. The public repository includes behavioral tests for parsing, independent client processes, output contracts and evidence verification. Both images build successfully and resolve anonymously from the registry. Native Radeon evaluation passed 10 OCR illustrations, 32 synthetic degraded OCR cases and all 10 RAG sample questions with exact citation sets. The required Instinct/container environment and private evaluation set have not been tested. Public examples and synthetic cases are development checks, not a claim about hidden-set scores.

KGB

AMD ROCmQwen3-VLHuggingFace Hub

VisionForge OCR + Grounded RAG on AMD ROCm

Two offline AMD ROCm systems share this entry: VisionForge AI for multilingual visual OCR (Match 2) and Grounded RAG for document question answering (Match 3). Each has a separate public Docker image in the matching private image field. Both include pinned model weights and need no runtime model downloads or external inference APIs. Match 2 reads US and Chinese registration plates and road signs from PNG, JPEG and TIFF. A resident Qwen3-VL-2B-Instruct worker loads once; a thin Python CLI sends image requests over a local Unix socket and writes the required JSON text output. US jurisdiction banners are excluded; Chinese province symbols remain part of the registration. Direct ROCm code passed 17/17 own synthetic cases on an AMD Radeon Pro W7900D (16.01 s startup, 3.57 s for ten CLI calls, 4.790 GiB observed peak). GPU inference inside the final Docker image was not tested. Match 3 answers questions over PDF, Word tables, spreadsheets, CSV, logs, Python source, diagrams and photographed labels with exact file citations. When evidence is missing it returns an empty answer and no citations. Resident Qwen3-VL-8B and Qwen3-Embedding-0.6B models combine lexical and semantic retrieval, identifier links and a second evidence check. Direct GPU code passed 287/287 development questions with answers and exact citation sets, plus 47 Linux tests. Peak PyTorch reserved memory was 19.87 GiB. Full Docker plus AMD GPU launch remains unverified because the free notebook has no container runtime. Both images preserve the required ROCm base. The tests use development or self-generated cases and do not establish an official hidden-set score.

Lord

AMD ROCmAMD Developer CloudQwen3-VL+1

RIOSE

RIOSE is a livestock simulation platform built for modern cattle operations, enabling producers and agribusiness teams to compare management scenarios, evaluate modeled impacts, and test decisions reproducibly.

RIOSE

AMD ROCmOpenAIrest api+2

AMD Hardware Advisor

A RAG-powered AI assistant that answers AMD hardware questions using official documentation, provides verifiable citations, compares products side by side, and recommends hardware for AI workloads.

CortexCrew

AMD ROCmHuggingFace HubStreamlit+1

Clearsheet: discharge sheets patients can follow

Turns a hospital discharge summary into instructions a patient can follow, in their own language. Gemma 3 27B on an AMD Instinct MI300X drafts; code checks every number and quote against the source; a nurse releases flagged lines.

Innova Ottawa

AMD ROCmGemmaDeepmind+1

Homeward: discharge teach-back in 12 languages

For discharge nurses and patients: Homeward rewrites discharge instructions in the patient's language with every dose and date checked by code, hears the patient explain them back, and shows the nurse what to re-teach. Gemma 4, fine-tuned on an AMD GPU.

Solus

GemmaAMD ROCmClaude Code+2

AMD Multi-Modal AI Solutions

royek ini menghadirkan rangkaian solusi kecerdasan buatan multi-modal berkinerja tinggi yang dioptimalkan untuk ekosistem komputasi AMD ROCm dan PyTorch. Seluruh modul solusi dikemas ke dalam arsitektur container Docker terstandarisasi untuk menjamin portabilitas, efisiensi konsumsi memori GPU, dan kecepatan inferensi pada lingkungan produksi maupun pengujian otomatis. Solusi ini mengintegrasikan fondasi model multi-modal mutakhir untuk menangani pemrosesan data visual dan bahasa secara terpadu, dimulai dari pengenalan teks visual (OCR) berlatensi rendah dengan keluaran terstruktur format JSON, hingga kesiapan pipeline inferensi untuk integrasi beban kerja AI lanjutan sesuai parameter evaluasi teknis kompetisi.

SmartClass Solo

AMD Developer CloudAMD ROCmHuggingFace Spaces+5

Grounded Web Retriever

Ask a question in plain English and the agent returns one exact value plus the page it came from: a phone number, a release version, the name of a current office holder. The rule that shapes everything else: an answer is only returned if it appears on a page the agent actually fetched. A value the model remembers but cannot find on the web is thrown away. Each question gets two model calls. First, Qwen3-4B-Instruct plans the search: the answer type, the entity, two queries. Then keyless search (Brave, DuckDuckGo and Seznam, with Bing as a fallback) runs alongside a Wikidata lookup of the entity's official website. "Latest version" questions also go straight to endoflife.date, PyPI, npm and the project's own GitHub releases. The agent fetches the top pages, puts the official domain first, and follows contact or release links one hop. Typed extractors pull candidate values from text, tel: links and JSON-LD. The second model call reads the best excerpts and picks a value. The code then checks every candidate against the fetched pages and scores it by source, official domain, nearby question words and agreement between independent sites. Confidence is computed, not generated. On 16 held-out questions (software versions, official phone numbers, current CEOs, answers checked by hand), the agent answered 11 correctly. The same model with no web access answered 0: outdated versions, invented phone numbers. Those numbers come from Qwen3-1.7B on a laptop GPU; the 4B model in the container has not been measured yet, and nothing has been measured on AMD hardware. The container is built on the mandated ROCm base and keeps the model resident in a local server, so each per-question process is a thin client. No API keys anywhere. Mini Challenge 4 Image: the submission form has no Mini Challenge 4 Image field yet, so the image is listed here: ghcr.io/stuartyo/amd-match4:latest (public).

StuartYo

AMD ROCmQwen3HuggingFace Hub

T2V · Text to Vision

Describe a classifier in words, get a working vision model. An AMD TokenFactory agent reasons about each class; a diffusion model writes all 274,042 weights in 0.08 s. No training, no images.

Delta Team

HuggingFace HubAMD ROCmDeepSeek V3+3

ROCm RAG: Cite-or-Refuse Document QA

This project implements a retrieval-augmented generation (RAG) system for the AMD AI League Match 3 challenge, running entirely on AMD ROCm hardware. The system ingests a mixed corpus of PDF, DOCX, XLSX, CSV, TXT, and image files, applies structure-aware chunking, and builds a dense vector index using Qwen3 embedding models. At query time, it retrieves the most relevant chunks and generates answers with an instruction-tuned language model, returning exact file citations for every answer. A core design principle is cite-or-refuse: when retrieved evidence does not support an answer, the system refuses rather than hallucinating, which is critical for citation-accuracy scoring. The full pipeline is packaged as a single ROCm container image that completes startup and indexing within the required time limits. Both the container image and source code are public.

solodog

AMD ROCmHuggingFace Hub

QuestBridge: reviewable quest localisation

A standalone MIT game-localisation prototype with real open-model inference on AMD Ryzen, immutable game tokens, visible rejection, reviewable drafts and a verifiable cache.

QuestBridge

Qwen3HuggingFace Hubrest api

Local OCR and RAG on AMD ROCm

This project contains my submissions for Mini-Challenge 2 (OCR) and Mini-Challenge 3 (RAG) of the AMD AI League. Both run in Docker images built on the mandated ROCm PyTorch base image, with the model weights shipped inside the image, so no network access is needed at evaluation time. Both solutions share the same architecture: a long-running FastAPI server, started by the container, keeps Qwen3-VL-2B-Instruct loaded on the GPU, and app.py is a thin client that the harness calls once per image or question. The model is never reloaded between calls. For OCR, a short English prompt with few-shot examples reads licence plates and road signs, with an automatic denoising and contrast step for dark images. For RAG, a single indexing pass converts every file type (PDF, Word, Excel, CSV, text, logs, Python, and images through the vision model) into text, skipping unreadable, encrypted and unknown files without stopping. Retrieval uses IDF weighting, a filename bonus and rare-identifier links for multi-hop questions. Citations and refusals are enforced by code rather than left to the model: the answer must appear in a cited file, every number or identifier in the question must appear in the sources, withdrawn revisions are excluded, and a linked file is cited only when the question does not already provide its identifier.

Kerdosmedia

AMD ROCmAMD Developer CloudHuggingFace Hub+1

Intelligent-Web-Information-Retrieval

A web retrieval agent that navigates links and GET forms, renders JavaScript with Scrapling, reads tables and PDFs, resolves conflicting evidence, and answers with verified citations. The latest agent workflow uses the local Qwen model from checkpoint 5 through a resident Transformers server. No ChatGPT login or hosted LLM API is needed. The model is Qwen/Qwen3-4B-Instruct-2507 with Transformers 5.20.2, BF16 weights, SDPA attention, deterministic generation and a 128-token output cap. These match the checkpoint's model configuration. The HTTP server keeps the model loaded between CLI queries; the agent's planning, extraction and review prompts are from the newer project. Features: - Navigate on the starting website, follow relevant links and submit discovered GET search forms. - Extract structured text from HTML and tables; render sparse JavaScript pages with Scrapling Chromium. - Read selectable-text PDFs and cite their page numbers. - Validate source passages, review meaning, handle competing claims and abstain when evidence is insufficient. - Run 33 local and public-site evaluation cases with traces and reports.

Ahmed Sa3ed

AMD ROCmAMD Developer CloudHuggingFace Hub+1

30apps loaded