Ai React Native | AI Final Year Project Report

Complete Technical Paper Content — Ai React Native

1. Executive Abstract, Problem Formulation & Background

Meta AI is a research division of Meta (formerly Facebook) that develops artificial intelligence and augmented reality technologies. It has workspaces in Menlo Park, London, New York City, Paris, Seattle, Pittsburgh, Tel Aviv, and Montreal as of 2025.

Llama is an LLM released in February 2023. As of January 2026, the most recent release is the Llama 4.

Meta Superintelligence Labs (MSL) is an American artificial intelligence division of Meta Platforms, headquartered in Menlo Park, California, founded in June 2025. The division focuses on research and development in the field of artificial superintelligence. MSL produces the Muse family of generative AI models, including the Muse AI agent, the Muse Spark large language model (LLM), Muse Image and Muse Video. MSL was preceded by Facebook Artificial Intelligence Research (FAIR), directed by Yann LeCun. In 2017 FAIR released PyTorch, an open source deep learning framework later used by ChatGPT, Tesla Autopilot, Uber's Pyro, and Hugging Face's Transformers. When Meta rebranded in 2021, FAIR was renamed Fundamental Artificial Intelligence Research, and Meta AI was formed. From 2023 to 2025, Meta released four iterations of the Llama LLM. In June 2025, Meta began developing a larger model codenamed Behemoth, reportedly motivated by CEO Mark Zuckerberg's displeasure with Llama 4. Meta spent more than $14 billion to acquire a 49% non-voting stake in Scale AI, a data annotation and language model benchmarking company. Scale AI's CEO Alexandr Wang moved to Chief AI Officer at Meta, leading MSL. Former GitHub CEO Nat Friedman joined as head of product. Meta had also unsuccessfully sought the acquisition of Safe Superintelligence Inc., Thinking Machines Lab, and Perplexity AI. In November 2025, LeCun left Meta to found Advanced Machine Intelligence Labs. Muse Spark was released in July 2026, achieving a top five result in benchmarks by Artificial Analysis.

In June 2025, Bloomberg News reported that Mark Zuckerberg, the chief executive of Meta Platforms, had expressed displeasure at Llama 4, the company's large language model released in April, tasking employees to work overtime. In response, Meta began internally developing Behemoth, a larger model set to be more sophisticated than offerings from OpenAI, Anthropic, and Google. According to The Wall Street Journal, amid concerns from Meta's leadership over Behemoth's capabilities, the company delayed the release of the model. The decision to delay Behemoth led Zuckerberg to involve himself closely with Meta's AI efforts, starting a WhatsApp group chat with senior leadership to recruit researchers. According to Bloomberg News, Zuckerberg set a goal to hire approximately fifty people to staff a firm to achieve artificial general intelligence. That month, Zuckerberg sought to invest several billion dollars into Scale AI and hire its chief executive and founder, Alexandr Wang. In addition, he had personally recruited researchers at his homes in Lake Tahoe and Palo Alto, California; The New York Times later reported that Zuckerberg had offered compensation packages valued between US$1 to US$100 million to employees at OpenAI and Google. Days later, Meta announced that it was investing US$14.3 billion into Scale AI, an intentionally muted role despite hiring Wang in order to avoid scrutiny from the Federal Trade Commission amid an impending decision from judge James Boasberg in FTC v. Meta (2020). According to The Information, Zuckerberg was willing to provide US$5 billion, though Wang countered with US$20 billion. In order to fund the tentative firm, Meta implemented advertisements in WhatsApp. The Information later reported that Meta was discussing hiring Nat Friedman, the former chief executive of GitHub, and the businessman and investor Daniel Gross, and acquiring their venture capital firm, NFDG. According to CNBC, Meta had sought to acquire Safe Superintelligence Inc., but its CEO Ilya Sutskever refused the acquisition. Additionally, Zuckerberg privately discussed acquiring Thinking Machines Lab and Perplexity AI, though the deals fell through over disputes concerning prices and strategy. Days later, The Verge reported that Gross and Friedman would report directly beneath Wang. Zuckerberg assumed a dominant role in hiring employees, though his efforts faced complications from researchers who expressed skepticism at Meta's artificial intelligence, uncertainty over internal restructuring, and a perceived strategic conflict with Meta's vice president for artificial intelligence, Yann LeCun. Additionally, several researchers were surprised to receive messages from Zuckerberg, including one person who, believing a message they received to be a hoax, did not respond for several days. On June 30, Zuckerberg announced that he was establishing Meta Superintelligence Labs with Wang serving as chief AI officer and Friedman leading work on AI products. Meta AI (formerly Fundamental Artificial Intelligence Research) and several other divisions, including a new team called TBD Lab, dedicated to "developing the next generation" of Meta's large language models, were placed beneath Meta Superintelligence Labs. In an internal memo, Zuckerberg named eleven employees the company had hired. Zuckerberg's efforts forced other AI company executives, including Microsoft's Satya Nadella and OpenAI's Sam Altman, to attract researchers themselves. In July, Gross joined Superintelligence Labs as Friedman's counterpart. That month, The New York Times reported that Superintelligence Labs executives had discussed a proprietary AI model. In August, Meta restructured Meta Superintelligence Labs into four subgroups. On November 20, 2025, Yann LeCun left Meta's chief AI scientist role to start a new firm, Advanced Machine Intelligence Labs.

TBD Lab: Led by Wang, this team manages Meta's large language models; FAIR: An artificial intelligence research team; Products and Applied Research: Led by Friedman, this group focuses on consumer integration; MSL Infra: Led by Aparna Ramani, this team develops and maintains the infrastructure required to sustain artificial intelligence models;

Llama is an LLM released in February 2023. As of January 2026, the most recent release is the Llama 4.

2. Related Work, Literature Survey & Algorithmic Foundations

React Native AI A collection of on-device AI primitives for React Native with first-class Vercel AI SDK support. Run AI models directly on users' devices for privacy-preserving, low-latency inference without server costs. Features 🚀 Instant AI - Use built-in system models immediately without downloads 🔒 Privacy-first - Inference runs on-device, so data stays local. Cloud models are opt-in 🎯 Vercel AI SDK compatible - Use generateText , streamText , embed and other AI SDK functions with local models 🎨 Beyond text - Embeddings, transcription, speech synthesis, image and video generation, and vision models AI SDK Compatibility React Native AI AI SDK 0.11 and below v5 0.12 v6 0.13 and above v7 DevTools The AI SDK Profiler plugin captures OpenTelemetry spans from Vercel AI SDK requests and surfaces them in Rozenite DevTools. DevTools are runtime agnostic, so they work with on-device and remote runtimes. npm install @react-native-ai/dev-tools Rozenite must be installed and enabled in your app. See the Rozenite getting started guide . If the AI SDK Profiler panel stays empty after you send a message, close the React Native DevTools window and open a new one. A stale debugger session can keep the panel mounted without receiving telemetry. Available Providers Provider Built-in Platforms Runtime Description Apple ✅ Yes iOS Apple Apple Foundation Models, embeddings, transcription, speech ADK ✅ Yes Android Google ADK On-device Gemini Nano, with optional cloud Gemini Core AI ❌ No iOS Core AI Text, image, video, speech, embedding and vision models Llama ❌ No iOS, Android llama.rn Run GGUF models via llama.rn Apple Native integration with Apple's on-device AI capabilities. Built-in - no model downloads required, uses system models. Text Generation - Apple Foundation Models for chat and completion Embeddings - NLContextualEmbedding for 512-dimensional semantic vectors Transcription - SpeechAnalyzer for fast, accurate speech-to-text Speech Synthesis - AVSpeechSynthesizer for natural text-to-speech with system voices Installation npm install @react-native-ai/apple No additional linking needed, works immediately on iOS devices (autolinked). Usage import { apple } from '@react-native-ai/apple' import { generateText , embed , transcribe , generateSpeech as speech , } from 'ai' // Text generation with Apple Intelligence const { text } = await generateText ( { model : apple ( ) , prompt : 'Explain quantum computing' , } ) // Generate embeddings const { embedding } = await embed ( { model : apple . embeddingModel ( ) , value : 'Hello world' , } ) // Transcribe audio const { text : transcript } = await transcribe ( { model : apple . transcriptionModel ( ) , audio : audioBuffer , } ) // Text-to-speech const { audio } = await speech ( { model : apple . speechModel ( ) , text : 'Hello from Apple!' , } ) Availability Feature iOS Version Additional Requirements Text Generation iOS 26+ Apple Intelligence device Embeddings iOS 17+ - Transcription iOS 26+ - Speech Synthesis iOS 13+ iOS 17+ for Personal Voice See the Apple documentation for detailed setup and usage guides. Google ADK Run Gemini Nano on-device on Android through Google's Agent Development Kit . Built-in - Gemini Nano is provisioned by the system via AICore, so there are no model files to bundle and no API key. Cloud Gemini is available as an option. Text Generation - generateText and streamText with on-device Gemini Nano Tool Calling - Native ADK agent loop with JavaScript tool executors Cloud Gemini - Optional cloud inference via ADK LlmAgent when you need a larger model Installation npm install @react-native-ai/adk Important Requires Android minSdkVersion 26 and a physical device with AICore support (Android 14+) for Gemini Nano. Your app also needs packaging excludes for duplicate META-INF files, see the getting started guide . Usage import { adk } from '@react-native-ai/adk' import { generateText } from 'ai' if ( ! ( await adk . isAvailable ( ) ) ) { throw new Error ( 'Gemini Nano is not available on this device' ) } await adk . prepareNano ( ) const { text } = await generateText ( { model : adk ( ) , prompt : 'Explain quantum computing' , } ) See the ADK documentation for device checks, cloud Gemini, and tool calling. Core AI Run models exported with Apple's coreai-models on Apple's Core AI runtime, on iOS 27 devices. Your app downloads or bundles the models, and every task has a hosted export to start from, except raw function calls, which run your own .aimodel . Text Generation - Qwen3 and other LLM exports, with structured output, tool calling and reasoning Image Prompts - Qwen3-VL answers questions about an image Image and Video Generation - FLUX.2 diffusion pipelines, and Wan 2.1 video on macOS (it does not fit in iPhone memory yet) Transcription - Parakeet TDT, for files and live audio Embeddings - CLIP text and image embeddings in one space Vision - Object detection, segmentation, video tracking, super-resolution, classification and depth Any .aimodel - Raw function calls with tensors and images Installation npm install @react-native-ai/core-ai @react-native-ai/apple ai Important Requires Xcode 27, an iOS 27 deployment target and a physical device: the iOS Simulator has no Core AI framework. With Expo, add the @react-native-ai/core-ai config plugin. Usage import { coreAI } from '@react-native-ai/core-ai' import { generateImage , streamText , transcribe } from 'ai' // Text generation with an exported Qwen3 const result = streamText ( { model : coreAI . languageModel ( { id : 'qwen3-0.6b' , source : qwen3 , } ) , prompt : 'Explain Core AI in one sentence.' , } ) for await ( const delta of result . textStream ) { console . log ( delta ) } // Image generation with FLUX.2 klein const { image } = await generateImage ( { model : coreAI . imageModel ( { id : 'flux2-klein-4b' , source : flux , } ) , prompt : 'a red apple on a wooden table' , } ) // Transcription with Parakeet TDT const { text } = await transcribe ( { model : coreAI . transcriptionModel ( { id : 'parakeet' , source : parakeet , } ) , audio : recording , } ) See the Core AI documentation for native setup, every task and the hosted models. Llama Run any GGUF model on-device using llama.rn . Requires download - models are downloaded from HuggingFace. Supported Features Feature Method Description Text Generation llama.languageModel() Chat, completion, streaming, reasoning models Embeddings llama.embeddingModel() Text embeddings for RAG and similarity search Speech llama.speechModel() Text-to-speech with vocoder models Installation npm install @react-native-ai/llama llama.rn react-native-blob-util Usage import { downloadModel , llama } from '@react-native-ai/llama' import { generateText } from 'ai' // Download from HuggingFace (Model ID format: "owner/repo/filename.gguf") const modelPath = await downloadModel ( 'bartowski/Qwen_Qwen3.5-2B-GGUF/Qwen_Qwen3.5-2B-Q4_K_M.gguf' , ( progress ) => console . log ( `Downloading: ${ progress . percentage } %` ) ) // Create model instance const model = llama . languageModel ( modelPath ) // Initialize model (loads into memory) await model . prepare ( ) // Generate text const { text } = await generateText ( { model , messages : [ { role : 'system' , content : 'You are a helpful assistant.' } , { role : 'user' , content : 'Write a haiku about coding.' } , ] , } ) // Cleanup when done await model . unload ( ) Model Compatibility Any GGUF model from HuggingFace can be used. Use the format owner/repo/filename.gguf as the model ID. Popular choices include: bartowski/Qwen_Qwen3.5-2B-GGUF/Qwen_Qwen3.5-2B-Q4_K_M.gguf bartowski/Llama-3.2-3B-Instruct-GGUF/Llama-3.2-3B-Instruct-Q4_K_M.gguf Qwen/Qwen2.5-1.5B-Instruct-GGUF/qwen2.5-1.5b-instruct-q4_k_m.gguf 📚 View full Llama documentation → JSON UI @react-native-ai/json-ui lets a tool-calling model build React Native UI. The model calls tools that add, update, and delete nodes in a JSON spec, and GenerativeUIView renders it. It's designed for small on-device models with limited context. npm install @react-native-ai/json-ui See the JSON UI documentation . Documentation Guides and API reference are available at react-native-ai.dev . Contributing Read the contribution guidelines before contributing. Agent skills This repository provides agent skills to help you integrate and use the packages. You can easily install them with: npx skills add https://github.com/callstackincubator/react-native-ai --skill react-native-ai-skills or manually by copying the skills/ directory into your agent's skills directory (for example .cursor/skills/ ). Made with ❤️ at Callstack react-native-ai is an open source project and will always remain free to use. If you think it's cool, please star it 🌟. Callstack is a group of React and React Native geeks, contact us at hello@callstack.com if you need any help with these or just want to say hi! Made with create-react-native-library

A collection of on-device AI primitives for React Native with first-class Vercel AI SDK support. Run AI models directly on users' devices for privacy-preserving, low-latency inference without server costs.

The AI SDK Profiler plugin captures OpenTelemetry spans from Vercel AI SDK requests and surfaces them in Rozenite DevTools. DevTools are runtime agnostic, so they work with on-device and remote runtimes.

Rozenite must be installed and enabled in your app. See the Rozenite getting started guide .

If the AI SDK Profiler panel stays empty after you send a message, close the React Native DevTools window and open a new one. A stale debugger session can keep the panel mounted without receiving telemetry.

Native integration with Apple's on-device AI capabilities. Built-in - no model downloads required, uses system models.

3. Dataset Ingestion, Preprocessing & Stochastic Augmentations

import { apple } from '@react-native-ai/apple' import { generateText , embed , transcribe , generateSpeech as speech , } from 'ai' // Text generation with Apple Intelligence const { text } = await generateText ( { model : apple ( ) , prompt : 'Explain quantum computing' , } ) // Generate embeddings const { embedding } = await embed ( { model : apple . embeddingModel ( ) , value : 'Hello world' , } ) // Transcribe audio const { text : transcript } = await transcribe ( { model : apple . transcriptionModel ( ) , audio : audioBuffer , } ) // Text-to-speech const { audio } = await speech ( { model : apple . speechModel ( ) , text : 'Hello from Apple!' , } )

Run Gemini Nano on-device on Android through Google's Agent Development Kit . Built-in - Gemini Nano is provisioned by the system via AICore, so there are no model files to bundle and no API key. Cloud Gemini is available as an option.

Requires Android minSdkVersion 26 and a physical device with AICore support (Android 14+) for Gemini Nano. Your app also needs packaging excludes for duplicate META-INF files, see the getting started guide .

import { adk } from '@react-native-ai/adk' import { generateText } from 'ai' if ( ! ( await adk . isAvailable ( ) ) ) { throw new Error ( 'Gemini Nano is not available on this device' ) } await adk . prepareNano ( ) const { text } = await generateText ( { model : adk ( ) , prompt : 'Explain quantum computing' , } )

Run models exported with Apple's coreai-models on Apple's Core AI runtime, on iOS 27 devices. Your app downloads or bundles the models, and every task has a hosted export to start from, except raw function calls, which run your own .aimodel .

Requires Xcode 27, an iOS 27 deployment target and a physical device: the iOS Simulator has no Core AI framework. With Expo, add the @react-native-ai/core-ai config plugin.

4. Neural Network Backbone, Layer Specifications & Loss Formulations

import { coreAI } from '@react-native-ai/core-ai' import { generateImage , streamText , transcribe } from 'ai' // Text generation with an exported Qwen3 const result = streamText ( { model : coreAI . languageModel ( { id : 'qwen3-0.6b' , source : qwen3 , } ) , prompt : 'Explain Core AI in one sentence.' , } ) for await ( const delta of result . textStream ) { console . log ( delta ) } // Image generation with FLUX.2 klein const { image } = await generateImage ( { model : coreAI . imageModel ( { id : 'flux2-klein-4b' , source : flux , } ) , prompt : 'a red apple on a wooden table' , } ) // Transcription with Parakeet TDT const { text } = await transcribe ( { model : coreAI . transcriptionModel ( { id : 'parakeet' , source : parakeet , } ) , audio : recording , } )

Run any GGUF model on-device using llama.rn . Requires download - models are downloaded from HuggingFace.

import { downloadModel , llama } from '@react-native-ai/llama' import { generateText } from 'ai' // Download from HuggingFace (Model ID format: "owner/repo/filename.gguf") const modelPath = await downloadModel ( 'bartowski/Qwen_Qwen3.5-2B-GGUF/Qwen_Qwen3.5-2B-Q4_K_M.gguf' , ( progress ) => console . log ( `Downloading: ${ progress . percentage } %` ) ) // Create model instance const model = llama . languageModel ( modelPath ) // Initialize model (loads into memory) await model . prepare ( ) // Generate text const { text } = await generateText ( { model , messages : [ { role : 'system' , content : 'You are a helpful assistant.' } , { role : 'user' , content : 'Write a haiku about coding.' } , ] , } ) // Cleanup when done await model . unload ( )

Any GGUF model from HuggingFace can be used. Use the format owner/repo/filename.gguf as the model ID. Popular choices include:

@react-native-ai/json-ui lets a tool-calling model build React Native UI. The model calls tools that add, update, and delete nodes in a JSON spec, and GenerativeUIView renders it. It's designed for small on-device models with limited context.

This repository provides agent skills to help you integrate and use the packages. You can easily install them with:

5. Training Protocol, Convergence Behavior & Experimental Evaluation

npx skills add https://github.com/callstackincubator/react-native-ai --skill react-native-ai-skills

or manually by copying the skills/ directory into your agent's skills directory (for example .cursor/skills/ ).

react-native-ai is an open source project and will always remain free to use. If you think it's cool, please star it 🌟.

Callstack is a group of React and React Native geeks, contact us at hello@callstack.com if you need any help with these or just want to say hi!

react-native-caffe2 Bring deep learning to mobile in 2 minutes 📱🥂 react-native-caffe2 is a library made in order to bring deep learning to mobile very fast. The main vision is to implement heavily used deep learning solution (image classification, style transfer, text generation, ..) with high-level methods in React Native to make possible for anyone to create fully fonctional deep learning applications for Android and iOS very fast. Important notes ⚠️ The current version is only experimental No support for Android for technical reasons from React Native (NDK required for React Native is version 10, for Caffe2 is superior than version 13) Support ios simulator only (no real device), device support coming soon Getting started [iOS] 📱 $ npm install react-native-caffe2 --save - Mostly automatic installation $ react-native link react-native-caffe2 - Manual installation In XCode, in the project navigator, right click Libraries ➜ Add Files to [your project's name] Go to node_modules ➜ react-native-caffe2 and add RNCaffe2.xcodeproj In XCode, in the project navigator, select your project. Add libRNCaffe2.a to your project's Build Phases ➜ Link Binary With Libraries Run your project ( Cmd+R )< Additional requirements Frameworks used have to be manually linked to your project. In XCode, go to File option, click on Add Files to , pick Copy items of needed option Go to ../node_modules/react-native-caffe2/ios/Frameworks/ Add opencv2.framework In project navigator, right click on Project, go to Build Phases ➜ Link Binary With Libraries Add the following frameworks : CoreMedia.framework AssetsLibrary.framework CoreVideo.framework Usage 🚀 One function is currently implemented: classifyImage The arguments are: imageName : Name of the image to classify initModel : protobuf file for initialization model predictModel : protobuf file for predicition model classes : Array of classes to for prediction callback : Callback method called with predicted class Unfortunately, all files (protobuf, image) must be recognized from ios code. Therefore every asset has to be put in the xcode project. In order to add an asset to the xcode project, you simply have to add the file in from the XCode IDE. index.ios.js import React , { Component } from 'react' ; import { AppRegistry , StyleSheet , Text , View } from 'react-native' ; import RNCaffe2 from 'react-native-caffe2' ; export default class toast extends Component { constructor ( props ) { super ( props ) ; this . state = { label : 'react-native-caffe2' } ; } render ( ) { var json = require ( './classes' ) ; RNCaffe2 . classifyImage ( 'dog.jpg' , 'exec_net' , 'predict_net' , json . classes , ( error , label ) => { if ( error ) { console . error ( error ) ; } else { this . setState ( { label : label } ) ; } } ) ; return ( < View style = { styles . container } > < Text style = { styles . welcome } > { this . state . label } </ Text > </ View > ) ; } } const styles = StyleSheet . create ( { container : { flex : 1 , justifyContent : 'center' , alignItems : 'center' , backgroundColor : '#F5FCFF' , } , welcome : { fontSize : 20 , textAlign : 'center' , margin : 10 , } , } ) ; AppRegistry . registerComponent ( 'appName' , ( ) => appName ) ; Methods coming : Classify from camera Classify general file Raw prediction (provide tensor and get tensor) TODO 📝 Add high-level methods Add support for Android Make the library less disk consuming Allow user to provide file path to method without putting the file in xcode assets Find a way to not have to do the Additional Requirement stuff

react-native-caffe2 is a library made in order to bring deep learning to mobile very fast. The main vision is to implement heavily used deep learning solution (image classification, style transfer, text generation, ..) with high-level methods in React Native to make possible for anyone to create fully fonctional deep learning applications for Android and iOS very fast.