Top Builders

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

LLaVA: Large Language and Vision Assistant

LLaVA represents a novel end-to-end trained large multimodal model that combines a vision encoder and Vicuna for general-purpose visual and language understanding, achieving impressive chat capabilities mimicking spirits of the multimodal GPT-4 and setting a new state-of-the-art accuracy on Science QA.

General
Relese dateNovember 20, 2023
Repositoryhttps://github.com/haotian-liu/LLaVA
TypeMultimodal Language and Vision Model

What is LLaVA?

Visual Instruction Tuning: LLaVA, short for Large Language-and-Vision Assistant, represents a significant leap in multimodal AI models.

With a focus on visual instruction tuning, LLaVA has been engineered to rival the capabilities of GPT-4V, demonstrating its exceptional prowess in understanding both language and vision. This state-of-the-art model excels in tasks ranging from impressive chatbot interactions to setting a new standard in science question-answering accuracy, achieving a remarkable 92.53%. With LLaVA's innovative approach to instruction-following data and the effective combination of vision and language models, it promises a versatile solution for diverse applications, marking a significant milestone in the field of multimodal AI.

LLaVA Tutorials

LLaVA Libraries

A curated list of libraries and technologies to help you build great projects with 'technology'.

Edit on GitHub

LLavA AI technology page Hackathon projects

Discover innovative solutions crafted with LLavA AI technology page, developed by our community members during our engaging hackathons.

Parallax

Parallax

PARALLAX is a GenAI-native cyber-fraud investigation platform for banks, fraud teams, SOC analysts, and cyber cells responding to APK-led financial crime. It analyzes suspicious Android applications and related fraud evidence to identify banking-malware behavior such as accessibility abuse, overlay attacks, SMS or notification interception, suspicious permissions, network indicators, and runtime compromise signals. PARALLAX converts these findings into structured evidence, confidence-scored claims, risk assessments, and investigation artifacts. The system is designed around an evidence-first workflow: ingest a suspicious APK, run static and dynamic analysis, extract IOCs and observations, snapshot the evidence bundle, coordinate specialist AI agents, surface challenges or disagreements, and produce a final human-reviewable action packet. Instead of treating fraud review as a static report, PARALLAX models it as a live investigation. Specialist agents can assess device compromise, transaction traces, mule-account patterns, evidence quality, liability context, legal evidence requirements, and final recommendations. The result is an auditable reasoning trail that helps teams understand not just what the system concluded, but why. PARALLAX keeps humans in control. Its outputs are designed to support analysts and bank officers with clear evidence, unresolved questions, recommended next steps, and exportable documentation for escalation or cyber-cell reporting.

ReachIQ AI - Generative Growth as a Service

ReachIQ AI - Generative Growth as a Service

ReachIQ AI is a Generative Growth as a Service platform built on Band's multi-agent coordination framework. Four specialized agents -BrainAgent, AnalyzerAgent, MonitorAgent, and DistributionAgent - collaborate autonomously to grow your YouTube channel. The system runs a fully autonomous workflow: Monitor Agent fetches real YouTube Analytics to identify your weakest-performing video, Analyzer Agent scores it across 6 dimensions and runs a 3-pass recursive self-improvement loop to optimize the title, metadata, and thumbnail, then Distribution Agent generates platform-specific promotional posts for YouTube Community, Facebook, LinkedIn, Reddit, Quora, and Twitter simultaneously. Built with Groq Llama 3.3 70B for sub-second AI reasoning, Mem0 and Qdrant for persistent channel memory that learns over time, Langfuse for full observability, and custom security guardrails blocking prompt injection. ReachIQ AI targets 45 million YouTube creators, 90% of whom have fewer than 10,000 subscribers due to a lack of intelligent growth infrastructure. The platform saves 20+ hours per week per creator and runs on a 100% free technology stack. Live demo available at reachiq-ai.streamlit.app

Sentinel Generalist:

Sentinel Generalist:

Sentinel Generalist is a zero-shot agricultural intelligence system powered by a large Vision-Language Model (Qwen2.5-VL-7B-Instruct) running on an AMD MI300X GPU. Upload a single photograph of any plant — indoors, outdoors, healthy, or dying — and Sentinel performs 11 simultaneous agricultural analyses in real time, streaming its step-by-step reasoning trace like an expert agronomist thinking out loud. Reasoning trace — watch the AI examine shadows, soil texture, leaf color, and turgor in real time Species identification with confidence scores (50,000+ species, zero-shot) Geospatial inference — USDA hardiness zone, latitude, climate class deduced from shadows and soil Light assessment — estimated daily sun hours + adequacy check Watering analysis — visual turgor + soil moisture cues Nutrient deficiency — N-P-K + micronutrients with organic remedy doses Pest & disease — IPM-first diagnosis (organic → chemical → prevention) Companion planting — recommended neighbors + antagonists + placement instructions Harvest intelligence — days to harvest, visual cues, succession planting Seasonal planning — frost risk, crop rotation, next-season prep Beginner garden planner — "What should I plant right now?" with 7 curated plants for your zone and season Garden harvest preview — interactive top-down layout + AI-generated image of your garden at peak harvest Why AMD MI300X? The 192GB HBM3 VRAM loads the full 7B model with room to spare. ROCm 7.0 + gfx942 optimization delivers ~23% faster inference than comparable cloud APIs. Your plant photos never leave the Droplet — privacy-first by design. The demo differentiator: Tech stack: Qwen2.5-VL-7B-Instruct · AMD MI300X · ROCm 7.0 · HuggingFace Optimum-AMD · FastAPI · React 19 · GitHub Pages · NDJSON streaming · 100/100 Lighthouse · zero trackers · no CDNs

AgriSync

AgriSync

AgriSync is an agricultural intelligence platform for smallholder farmers across Africa's 54 nations. Two problems destroy farm income every season: 1. Crop disease goes undiagnosed — the nearest agronomist is hours away or unaffordable. A farmer watches their maize develop grey leaf spot and doesn't know whether to spray, wait, or replant. 2. Farmers sell at the wrong market — without real-time price data, a farmer in Nakuru may sell tomatoes locally for KES 20/kg when Nairobi is paying KES 31/kg that same morning. On 500 kg that's KES 5,500 lost to information asymmetry. AgriSync solves both in one flow. A farmer takes a photo of their crop leaf with any smartphone. Our two-stage vision pipeline — Llama-3.2-11B-Vision-Instruct as primary, with LLaVA-v1.5-7B (fine-tuned on plant diseases) as specialist fallback, both running on AMD MI300X — identifies the disease, severity level, and the specific PCPB-approved chemical available at the nearest agro-vet, with the price in KES. At the same time, our ArbitrageEngine agent queries crop prices across African market hubs, calculates net profit after transport costs, and recommends the single best market to sell at today. The OrchestratorAgent (Mistral-7B-Instruct) combines both outputs into a bilingual English and Swahili advisory. For farmers without a smartphone, the same advisory is delivered as a 160-character SMS via Africa's Talking API. The platform covers diseases across major African food crops: Tomato Late Blight, Maize Gray Leaf Spot, Cassava Mosaic Disease, Fall Armyworm, Groundnut Rosette, Rice Yellow Mottle Virus, Bean Angular Leaf Spot, Potato Late Blight, and more.