Gemma

Top Builders

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

Gemma

Gemma is a lightweight, open large language model (LLM) from Google, optimized for efficient AI applications. As part of the Google Gemma family, it uses a transformer-based architecture tailored for responsible and accessible AI usage. Developed as a foundational model, Gemma serves various basic language processing needs, including chatbots, content summarization, and multilingual support.

General
Relese date	February 2024
Author	Google DeepMind in collaboration with Google AI teams
Website	[Google AI Gemma]https://ai.google.dev/gemma
Repository	Google AI Developer Resources
Type	Open-source AI, transformer-based LLM

Key Features

Efficient Deployment: Available in parameter sizes like 2.5B and 7B, Gemma balances capability with efficiency, enabling deployments on both edge devices and cloud infrastructure.
Flexible Tuning Options: Offers pre-trained and instruction-tuned variants, allowing developers to optimize for specific use cases or deploy as-is.
Decoder-Only Transformer Architecture: Uses a streamlined decoder-only design, enabling Gemma 1 to process up to 8192 tokens in one pass for better handling of long-form text.
Safety and Accessibility Tools: Integrates responsible AI features, promoting transparency and safety in AI outputs.

Applications:

Chatbot Development: Optimized for conversational tasks, Gemma provides foundational capabilities for chatbot applications.
Summarization and Paraphrasing: Its pre-trained model structure makes it suitable for summarizing content across languages and contexts.
Multilingual Processing: Supports multilingual inputs, making it adaptable for global applications and translation services.

Get started building with Gemma:

Developers can quickly integrate Gemma into applications by accessing its model weights on Google AI Studio and Kaggle. The model’s lightweight design ensures that it can run efficiently on most hardware configurations, including mobile and edge devices. For optimal performance, utilize frameworks such as Keras or JAX to customize and deploy Gemma for your specific use case. Get started today by exploring the tools and resources available on the Google AI Gemma platform.

Edit on GitHub

Google Gemma AI technology Hackathon projects

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

Band Memory

Band Memory gives multi-agent systems persistent, shared memory so they stop starting every task with amnesia. The problem: AI agents coordinated through Band.ai handle complex workflows — planning, executing, reviewing — but when the session ends, everything they learned evaporates. The Planner forgets which architecture decisions worked. The Executor re-discovers conventions it already learned. The Reviewer can't reference past findings. Every task starts from zero. Band Memory wires Mimir — a battle-tested persistent memory engine (Rust, SQLite+FTS5, 23 MCP tools) — directly into Band agents. Three agents (Planner, Executor, Reviewer) coordinate through Band rooms and share a common memory backend. Each agent has custom tools (remember, recall, forget) that persist and retrieve context across sessions. In the demo, Session 1 starts cold: the Planner checks for past auth decisions and finds nothing, creates a plan from scratch, the Executor establishes conventions (bcrypt, JWT patterns), and the Reviewer stores findings. In Session 2, the user asks to add OAuth — the Planner instantly recalls the auth architecture, the Executor pulls up the exact conventions, and the Reviewer cross-references past findings. The team compounds knowledge every run. Built with the Band SDK (LangGraph adapter), Mimir MCP server, and GPT-4o for agent reasoning. The skill file in agents/memory_tools.py can be reused by any Band agent. Zero cloud dependency for memory — Mimir runs locally on SQLite. This is what Band agents are missing: memory that survives the session. Not just structured chat history, but searchable, decaying, confidence-scored knowledge that compounds across every interaction.

🦷 Dental Surgical Simulator

The simulator presents a stylized, professional interface featuring a 3D viewport of a human dental arch. Users select from various surgical missions, each requiring a specific sequence of surgical steps. Success depends on selecting the correct surgical instrument and interacting with the precise target tooth at the right time. Procedural Anatomy: Anatomically accurate tooth crowns and root clusters generated via math-based geometry. Multi-Mode Visualization: X-Ray Mode: Toggles transparency of soft tissues (gums, palate) to reveal bone and roots. Wireframe Mode: Provides structural visualization of dental anatomy. View Switching: Toggle between Full Arch, Upper Jaw, and Lower Jaw views. Dynamic Theming: Seamless transition between a professional high-contrast Dark Mode and a clean Light Mode. Interactive HUD: Real-time feedback through toasts, tooltips, and a crosshair for precise targeting. Advanced Missions: Includes complex procedures like Dental Implants, Root Canals, and Bone Grafting.

Verified News

**Verified News** is a real-time narrative verification platform that combats misinformation by autonomously fact-checking claims against live web data, social signals, community notes, and fact-checking databases. Where services like Ground News show *bias*, Verified News answers a harder question: **is it true?** The platform deploys a multi-agent swarm built on **CrewAI**, powered by **Bright Data's** full web data infrastructure (MCP Server, SERP API, Web Unlocker, Scraping Browser, Web Scraper API), with all reasoning running through **AI/ML API** and **Gemma 4 31B**. Verdicts are delivered as color-coded trust badges (🔵 Blue, 🟢 Green, 🟡 Yellow, 🟠 Orange, 🔴 Red, ⚪ Gray) with full evidentiary transparency. For **enterprises**, Verified News is two products in one: - ** Security & Compliance:** A threat intelligence layer that detects disinformation campaigns, false product recall rumors, fake executive statements, and leaked data narratives before they crater stock prices or trigger regulatory action. - ** GTM Intelligence:** A brand health and competitive intelligence platform that verifies market narratives and competitor claims, giving revenue teams accurate ground truth for messaging and positioning.

Transcript Evaluator

Interview scoring with one LLM is fast but fragile: models can over-score vague answers, miss follow-ups, or invent reasoning not supported by the transcript. Transcript Evaluator solves this with a two-agent architecture. Agent 1 batch-evaluates interview JSON against a unified behavioral/non-behavioral rubric and exports a structured PDF report. Agent 2 runs as a local review service powered by Bright Data MCP, receiving each case plus the primary evaluation and using MCP tools (search_engine, extract, and AI insight tools like web_data_chatgpt_ai_insights) to extract context and cross-check the first agent’s output. In a separate PDF feedback loop, Agent 2 sends the report back to Agent 1, which polls external AI perspectives through MCP and returns bias and calibration feedback so Agent 2 can revise before finalizing scores—reducing single-model hallucination and improving auditability for hiring teams.

TrapScan — AI Agent Trap Detector

TrapScan is a browser-native AI security tool designed to detect adversarial web content targeting AI agents in real time. As AI assistants increasingly browse websites, summarize information, and execute tasks autonomously, the web itself is becoming an attack surface. Attackers can embed hidden instructions inside webpages that are invisible to humans but fully readable by AI systems. These attacks can manipulate AI reasoning, inject malicious prompts, override safeguards, or trigger unauthorized actions. Inspired by Google DeepMind’s “AI Agent Traps” research, TrapScan implements detection for six major AI agent attack categories, including prompt injection, semantic manipulation, jailbreak attempts, hidden behavioral control patterns, and systemic adversarial traps. TrapScan combines fast local browser detection with AI-powered classification using Gemma 4 (gemma-4-26b-a4b-it). The Chrome extension scans HTML, CSS, metadata, JSON-LD schema, hidden DOM elements, and suspicious prompt patterns directly inside the browser. Suspicious findings are then analyzed by Gemma 4, which classifies threats, assigns risk scores, filters false positives, and explains attacks in plain English. The project includes: * A Manifest V3 Chrome extension * Real-time threat analysis UI * Browser risk indicators * Scan history and downloadable audit reports * A live Vercel-hosted web demo for instant testing TrapScan represents a new category of browser-native defense tooling focused specifically on protecting AI agents from manipulation on the open web.