# John O'Reilly

John O'Reilly's Blog

This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles.

## Exposing Confetti sessions and bookmarks to AI agents using Android AppFunctions

DevFeed: [Exposing Confetti sessions and bookmarks to AI agents using Android AppFunctions](<https://devfeed.tech/articles/exposing-confetti-sessions-and-bookmarks-to-ai-agents-using-android-appfunctions-111021.md>)

Original publisher: [Read original article](<https://johnoreilly.dev/posts/confetti-appfunctions/>)

Published: 2026-11-01T00:00:00Z

Content type: tutorial

Language: en

Sources: [John O'Reilly](<https://devfeed.tech/sources/john-o-reilly.md>)

Topics: [Android](<https://devfeed.tech/topics/android.md>), [AI Bots](<https://devfeed.tech/topics/ai-bots.md>), [agentic workflows](<https://devfeed.tech/topics/agentic-workflows.md>)

Tags: [a2ui](<https://devfeed.tech/tags/a2ui.md>), [actions](<https://devfeed.tech/tags/actions.md>), [adb](<https://devfeed.tech/tags/adb.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [android](<https://devfeed.tech/tags/android.md>), [appfunctions](<https://devfeed.tech/tags/appfunctions.md>), [bookmark](<https://devfeed.tech/tags/bookmark.md>), [debug](<https://devfeed.tech/tags/debug.md>), [demo](<https://devfeed.tech/tags/demo.md>), [dependencies](<https://devfeed.tech/tags/dependencies.md>), [droidcon](<https://devfeed.tech/tags/droidcon.md>), [function](<https://devfeed.tech/tags/function.md>), [functions](<https://devfeed.tech/tags/functions.md>), [gemini](<https://devfeed.tech/tags/gemini.md>), [gradle](<https://devfeed.tech/tags/gradle.md>), [mcp](<https://devfeed.tech/tags/mcp.md>), [sessions](<https://devfeed.tech/tags/sessions.md>)

### AI overview

Confetti exposes four Android AppFunctions--`getConference`, `listSessions`, `addBookmark` and `removeBookmark`--so on-device agents can answer questions about the selected conference and manage session bookmarks. The implementation uses AppFunctions `1.0.0-beta01`, KSP-generated service code and schema, and existing repositories; it requires `compileSdk` 37 and `targetSdk` 36 or later. Function and serialized-property KDoc descriptions are passed to the LLM, and the generated service and app metadata must be registered. In the described Gemini Agent Demo test, Gemini found sessions about agents using the conference and session functions, then bookmarked a session that appeared in Confetti. The author had not yet tested with Android MCP.

### Source excerpt

AppFunctions allow an Android app to expose functionality that can then be used by AI agents and assistants running on the device. With the recent announcement of Android MCP (which is powered by the AppFunctions API and allows Gemini to make use of functionality in your app) I thought it was a good time to look at adding AppFunctions support to the Confetti conference app. This article will outline the changes made to do that (included in this PR).

## On-device embeddings with EmbeddingGemma 2 in the Koog-based Confetti AI agent

DevFeed: [On-device embeddings with EmbeddingGemma 2 in the Koog-based Confetti AI agent](<https://devfeed.tech/articles/on-device-embeddings-with-embeddinggemma-2-in-the-koog-based-confetti-ai-agent-86407.md>)

Original publisher: [Read original article](<https://johnoreilly.dev/posts/confetti-embeddinggemma2/>)

Published: 2026-10-12T23:00:00Z

Content type: tutorial

Language: en

Sources: [John O'Reilly](<https://devfeed.tech/sources/john-o-reilly.md>)

Topics: [Embeddings](<https://devfeed.tech/topics/embeddings.md>), [On-device AI](<https://devfeed.tech/topics/on-device-ai.md>), [AI Agent](<https://devfeed.tech/topics/ai-agent.md>), [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [android](<https://devfeed.tech/tags/android.md>), [android-17](<https://devfeed.tech/tags/android-17.md>), [api](<https://devfeed.tech/tags/api.md>), [app](<https://devfeed.tech/tags/app.md>), [core](<https://devfeed.tech/tags/core.md>), [embedding](<https://devfeed.tech/tags/embedding.md>), [embedding-model](<https://devfeed.tech/tags/embedding-model.md>), [embeddinggemma](<https://devfeed.tech/tags/embeddinggemma.md>), [embeddinggemma-2](<https://devfeed.tech/tags/embeddinggemma-2.md>), [embeddings](<https://devfeed.tech/tags/embeddings.md>), [gemini](<https://devfeed.tech/tags/gemini.md>), [gemma](<https://devfeed.tech/tags/gemma.md>), [google](<https://devfeed.tech/tags/google.md>), [kmp](<https://devfeed.tech/tags/kmp.md>), [koog](<https://devfeed.tech/tags/koog.md>), [llm](<https://devfeed.tech/tags/llm.md>), [npu](<https://devfeed.tech/tags/npu.md>), [on-device](<https://devfeed.tech/tags/on-device.md>), [phone](<https://devfeed.tech/tags/phone.md>), [pixel](<https://devfeed.tech/tags/pixel.md>), [run](<https://devfeed.tech/tags/run.md>), [semantic](<https://devfeed.tech/tags/semantic.md>), [semantic-search](<https://devfeed.tech/tags/semantic-search.md>)

### AI overview

The article explains how a Koog-based Android app uses EmbeddingGemma 2 to compute semantic-search embeddings locally, first on the CPU and then on a Tensor G5 NPU. In the author's test, building an index for 63 conference sessions took about 19 seconds on the CPU and about 6 seconds on the NPU; reported native memory use was around 460 MB and 70 MB, respectively. It also describes device-specific model selection, local caching, and limitations such as NPU support being specific to the Tensor G5.

### Source excerpt

In an earlier article we added semantic search to the Koog-based AI agent in the Confetti Compose Multiplatform sample. That used Gemini both for the LLM driving the agent and for the embeddings behind the session search. Google recently announced EmbeddingGemma 2, a small embedding model designed to run on-device, and in this article we'll look at using it to compute those embeddings on the phone instead (with Gemini still driving the agent), first on the CPU and then on the Tensor G5 NPU in a Pixel 10 Pro XL.

## Moving from SKIE to Swift export in a Kotlin Multiplatform project

DevFeed: [Moving from SKIE to Swift export in a Kotlin Multiplatform project](<https://devfeed.tech/articles/moving-from-skie-to-swift-export-in-a-kotlin-multiplatform-project-84502.md>)

Original publisher: [Read original article](<https://johnoreilly.dev/posts/peopleinspace-swift-export/>)

Published: 2026-10-09T23:00:00Z

Content type: tutorial

Language: en

Sources: [John O'Reilly](<https://devfeed.tech/sources/john-o-reilly.md>)

Topics: [Swift](<https://devfeed.tech/topics/swift.md>), [compose-multiplatform](<https://devfeed.tech/topics/compose-multiplatform.md>), [SwiftUI](<https://devfeed.tech/topics/swiftui.md>), [compose-hot-reload](<https://devfeed.tech/topics/compose-hot-reload.md>)

Tags: [kmp](<https://devfeed.tech/tags/kmp.md>), [kotlin](<https://devfeed.tech/tags/kotlin.md>), [kotlin-multiplatform](<https://devfeed.tech/tags/kotlin-multiplatform.md>), [skie](<https://devfeed.tech/tags/skie.md>), [swift](<https://devfeed.tech/tags/swift.md>), [swift-export](<https://devfeed.tech/tags/swift-export.md>), [swift-interop](<https://devfeed.tech/tags/swift-interop.md>)

### AI overview

The article explains how the PeopleInSpace Kotlin Multiplatform sample moved its SwiftUI client from an Objective-C framework with SKIE to Kotlin's alpha Swift export. It details Gradle and Xcode setup, reducing generated bindings by controlling exported APIs and dependencies, and adapting Swift code for generated namespaces, sealed classes, and flows. The author reports that the generated Swift output fell from about 18,000 lines to about 2,600; Swift export remains in alpha.

### Source excerpt

The SwiftUI client in the PeopleInSpace Kotlin Multiplatform sample has, up to now, consumed the shared Kotlin code through a framework with a generated Objective-C API, along with SKIE to make that code nicer to use from Swift (for things like sealed classes and flows). Kotlin also has Swift export (currently in alpha) which generates Swift bindings directly, and in this article we're going to go through what was needed to move PeopleInSpace over to it. The changes are in this PR.

## Kotlin Multiplatform iOS build times: what's actually driving the cost

DevFeed: [Kotlin Multiplatform iOS build times: what's actually driving the cost](<https://devfeed.tech/articles/kotlin-multiplatform-ios-build-times-what-s-actually-driving-the-cost-64358.md>)

Original publisher: [Read original article](<https://johnoreilly.dev/posts/kmp-ios-build-times/>)

Published: 2026-10-02T23:00:00Z

Content type: article

Language: en

Sources: [John O'Reilly](<https://devfeed.tech/sources/john-o-reilly.md>)

Topics: [Kotlin](<https://devfeed.tech/topics/kotlin.md>), [build times](<https://devfeed.tech/topics/build-times.md>), [Gradle](<https://devfeed.tech/topics/gradle.md>)

Tags: [build](<https://devfeed.tech/tags/build.md>), [build-times](<https://devfeed.tech/tags/build-times.md>), [kmp](<https://devfeed.tech/tags/kmp.md>), [kotlin](<https://devfeed.tech/tags/kotlin.md>), [kotlin-multiplatform](<https://devfeed.tech/tags/kotlin-multiplatform.md>)

### AI overview

The article benchmarks Kotlin Multiplatform iOS build times against Swift, examining Kotlin versions, framework setup, exported API size, and Swift export. It finds that limiting the API exposed to Swift can substantially reduce incremental build times, while Kotlin version changes have little effect in these tests. Swift export is slower than the Objective-C-based framework in the measured scenarios. The results use generated code on one machine, so real-world savings depend on how much code can remain internal.

### Source excerpt

A recent article, The Hidden Cost Behind Kotlin Multiplatform on iOS, looked at what KMP does to the iOS edit/build/run feedback loop. It's a good question to ask and the article also comes with a benchmark harness (kmp-ios-benchmark) that compares a Swift-only setup against an equivalent KMP one, using generated code across clean, incremental, public API change, multi-module and release builds. The headline numbers were that a single internal change rebuilt in 3.19s for Swift vs 10.36s for KMP, and that the gap grew with module size (KMP adding ~4.2x more time per 100 files than Swift).

## Rendering AI agent generated UI using Jetpack Compose A2UI and Koog

DevFeed: [Rendering AI agent generated UI using Jetpack Compose A2UI and Koog](<https://devfeed.tech/articles/rendering-ai-agent-generated-ui-using-jetpack-compose-a2ui-and-koog-59228.md>)

Original publisher: [Read original article](<https://johnoreilly.dev/posts/a2ui-compose-koog/>)

Published: 2026-09-23T23:00:00Z

Content type: tutorial

Language: en

Sources: [John O'Reilly](<https://devfeed.tech/sources/john-o-reilly.md>)

Topics: [a2ui](<https://devfeed.tech/topics/a2ui.md>), [AI Agent](<https://devfeed.tech/topics/ai-agent.md>), [Jetpack Compose](<https://devfeed.tech/topics/jetpack-compose.md>), [multiplatform](<https://devfeed.tech/topics/multiplatform.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [Android](<https://devfeed.tech/topics/android.md>), [Kotlin](<https://devfeed.tech/topics/kotlin.md>), [JSON Schema](<https://devfeed.tech/topics/json-schema.md>)

Tags: [a2ui](<https://devfeed.tech/tags/a2ui.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agent](<https://devfeed.tech/tags/ai-agent.md>), [android](<https://devfeed.tech/tags/android.md>), [compose](<https://devfeed.tech/tags/compose.md>), [jetpack-compose](<https://devfeed.tech/tags/jetpack-compose.md>), [json](<https://devfeed.tech/tags/json.md>), [json-schema](<https://devfeed.tech/tags/json-schema.md>), [koog](<https://devfeed.tech/tags/koog.md>), [kotlin](<https://devfeed.tech/tags/kotlin.md>), [kotlin-multiplatform](<https://devfeed.tech/tags/kotlin-multiplatform.md>), [llms](<https://devfeed.tech/tags/llms.md>)

### AI overview

This tutorial explains how to render AI-agent-generated user interfaces with A2UI and Jetpack Compose in a Kotlin Multiplatform sample. It covers the Android-only Compose renderer, shared agent code, catalog-based components, A2UI message processing, and Google Gemini integration through Koog.

### Source excerpt

A2UI is a protocol that allows an AI agent to describe a user interface as JSON which the client then renders using its own native components (the agent can only make use of components from a "catalog" that the client provides). Google have just released Jetpack Compose support for A2UI (currently 1.0.0-alpha01) and in this article we're going to show how it can be used along with the Koog based AI agent in the ClimateTraceKMP Kotlin Multiplatform sample (I covered the initial Koog exploration in that project in an earlier article). Koog can work with several different LLMs and we're using Google Gemini in this case.

## Using the new Google Gen AI SDK for Kotlin

DevFeed: [Using the new Google Gen AI SDK for Kotlin](<https://devfeed.tech/articles/using-the-new-google-gen-ai-sdk-for-kotlin-25196.md>)

Original publisher: [Read original article](<https://johnoreilly.dev/posts/genai-kotlin-sdk/>)

Published: 2026-09-04T23:00:00Z

Content type: article

Language: en

Sources: [John O'Reilly](<https://devfeed.tech/sources/john-o-reilly.md>)

Topics: [SDK](<https://devfeed.tech/topics/sdk.md>), [Kotlin](<https://devfeed.tech/topics/kotlin.md>), [coroutines-flow](<https://devfeed.tech/topics/coroutines-flow.md>), [Firebase](<https://devfeed.tech/topics/firebase.md>), [firebase-ai-logic](<https://devfeed.tech/topics/firebase-ai-logic.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [ai-sdk](<https://devfeed.tech/tags/ai-sdk.md>), [apis](<https://devfeed.tech/tags/apis.md>), [base64](<https://devfeed.tech/tags/base64.md>), [basketball](<https://devfeed.tech/tags/basketball.md>), [client](<https://devfeed.tech/tags/client.md>), [coroutines](<https://devfeed.tech/tags/coroutines.md>), [firebase](<https://devfeed.tech/tags/firebase.md>), [gemini](<https://devfeed.tech/tags/gemini.md>), [gen-ai](<https://devfeed.tech/tags/gen-ai.md>), [google](<https://devfeed.tech/tags/google.md>), [koog](<https://devfeed.tech/tags/koog.md>), [kotlin](<https://devfeed.tech/tags/kotlin.md>), [library](<https://devfeed.tech/tags/library.md>), [rest](<https://devfeed.tech/tags/rest.md>), [sdk](<https://devfeed.tech/tags/sdk.md>), [sdks](<https://devfeed.tech/tags/sdks.md>)

### AI overview

Google's Gen AI SDK for Kotlin replaces FormAI's hand-built Gemini REST calls with an idiomatic client built around coroutines and Flow. The JVM tool processes video clips with two requests: a plain-text coaching critique and schema-constrained JSON containing timestamps for four swing or movement moments. The SDK accepts video bytes directly and handles base64 encoding; its text convenience method joins text parts from the first candidate while skipping parts marked as thinking. The excerpt says the client can read an API key from an environment variable or receive one explicitly, and that the tool passes a request timeout. It also notes that adding the SDK removes the module's separate HTTP client and mirrored request/response data classes. The excerpt cuts off during the structured-output explanation, so further implementation details are unavailable.

### Source excerpt

Google have just released version 1.0 of the Gen AI SDK for Kotlin, an idiomatic Kotlin client for the Gemini APIs. Up to now, calling Gemini from Kotlin has generally meant use of one of the Firebase SDKs, a community wrapper, an agent framework like Koog, or (as in the code below) just talking to the REST endpoint yourself. This is Google's own Kotlin library for it, built around coroutines and Flow.

## Running golf swing analysis on an Android device with a fine-tuned Gemma 4 model

DevFeed: [Running golf swing analysis on an Android device with a fine-tuned Gemma 4 model](<https://devfeed.tech/articles/running-golf-swing-analysis-on-an-android-device-with-a-fine-tuned-gemma-4-model-25193.md>)

Original publisher: [Read original article](<https://johnoreilly.dev/posts/formai-gemma4-ondevice/>)

Published: 2026-08-29T23:00:00Z

Content type: article

Language: en

Sources: [John O'Reilly](<https://devfeed.tech/sources/john-o-reilly.md>)

Topics: [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>), [Kotlin](<https://devfeed.tech/topics/kotlin.md>), [gemma](<https://devfeed.tech/topics/gemma.md>), [lora](<https://devfeed.tech/topics/lora.md>), [Android](<https://devfeed.tech/topics/android.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [android](<https://devfeed.tech/tags/android.md>), [cross-platform](<https://devfeed.tech/tags/cross-platform.md>), [feedback](<https://devfeed.tech/tags/feedback.md>), [form](<https://devfeed.tech/tags/form.md>), [gemini](<https://devfeed.tech/tags/gemini.md>), [gemma](<https://devfeed.tech/tags/gemma.md>), [gemma-4](<https://devfeed.tech/tags/gemma-4.md>), [golf](<https://devfeed.tech/tags/golf.md>), [kotlin](<https://devfeed.tech/tags/kotlin.md>), [kotlin-multiplatform](<https://devfeed.tech/tags/kotlin-multiplatform.md>), [litert](<https://devfeed.tech/tags/litert.md>), [model](<https://devfeed.tech/tags/model.md>), [multiplatform](<https://devfeed.tech/tags/multiplatform.md>), [running](<https://devfeed.tech/tags/running.md>)

### AI overview

FormAI adds an Android-only option for golf swing feedback that runs locally with a Gemma 4 model fine-tuned to imitate Gemini on this task. The approach uses Gemini as a teacher to generate training data from video clips: a standalone bulk tool makes separate calls for a plain-text critique and schema-constrained timestamps for setup, backswing top, contact and finish, because combining strict JSON with prose reportedly worsened the prose. The model is fine-tuned with LoRA and converted for LiteRT-LM; the excerpt says other app targets still fall back to cloud analysis. It contrasts this custom supplied model with prior platform-provided models, where the developer has less control over what the model knows. The excerpt does not provide the conversion details or practical cost results promised by the article.

### Source excerpt

FormAI is a Kotlin Multiplatform app that analyses a video of your golf swing, basketball shot or running form and gives you coaching feedback. Up to now that has always meant uploading the video to Gemini and getting the response back over the network. We've added an option to do the golf swing analysis entirely on an Android device instead, using a small Gemma 4 model that we fine-tuned to imitate Gemini for that one task. LiteRT-LM, the runtime we use for this, is itself cross platform (Android, iOS, desktop and web), but we've only wired up the Android side so far, so this path lives in androidMain and the other targets report it as unavailable and fall back to the cloud.

## Engineering log: fine-tuning Gemma 4 E4B with LoRA to bring FormAI's coaching on-device

DevFeed: [Engineering log: fine-tuning Gemma 4 E4B with LoRA to bring FormAI's coaching on-device](<https://devfeed.tech/articles/engineering-log-fine-tuning-gemma-4-e4b-with-lora-to-bring-formai-s-coaching-on-device-25192.md>)

Original publisher: [Read original article](<https://johnoreilly.dev/posts/formai-gemma4-lora/>)

Published: 2026-08-28T23:00:00Z

Content type: tutorial

Language: en

Sources: [John O'Reilly](<https://devfeed.tech/sources/john-o-reilly.md>)

Topics: [lora](<https://devfeed.tech/topics/lora.md>), [gemma](<https://devfeed.tech/topics/gemma.md>), [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>), [On-device AI](<https://devfeed.tech/topics/on-device-ai.md>), [Enhancing AI model performance](<https://devfeed.tech/topics/enhancing-ai-model-performance.md>), [ai-experiments](<https://devfeed.tech/topics/ai-experiments.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [analysis](<https://devfeed.tech/tags/analysis.md>), [android](<https://devfeed.tech/tags/android.md>), [api](<https://devfeed.tech/tags/api.md>), [app](<https://devfeed.tech/tags/app.md>), [basketball](<https://devfeed.tech/tags/basketball.md>), [behaviour](<https://devfeed.tech/tags/behaviour.md>), [claude](<https://devfeed.tech/tags/claude.md>), [claude-code](<https://devfeed.tech/tags/claude-code.md>), [clip](<https://devfeed.tech/tags/clip.md>), [coaching](<https://devfeed.tech/tags/coaching.md>), [data-quality](<https://devfeed.tech/tags/data-quality.md>), [device](<https://devfeed.tech/tags/device.md>), [engineering](<https://devfeed.tech/tags/engineering.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [gemma](<https://devfeed.tech/tags/gemma.md>), [gemma-4](<https://devfeed.tech/tags/gemma-4.md>), [lora](<https://devfeed.tech/tags/lora.md>), [measurements](<https://devfeed.tech/tags/measurements.md>), [on-device](<https://devfeed.tech/tags/on-device.md>)

### AI overview

FormAI's prototype explores moving sports-form coaching from Gemini's cloud API to on-device inference by fine-tuning Gemma 4 E4B (4.5B effective parameters) with LoRA. It uses Gemini as a teacher on rights-cleared Pexels golf clips, filters clips that cannot support a critique, extracts still frames for training, then merges, quantizes, and converts the model to LiteRT-LM for Android. The excerpt reports that a close-up clip showing only the ball and club impact produced a cautious refusal to assess stance, grip, or swing arc; it identifies such responses as a data-quality concern. This is an AI-generated account of a prototype, and the excerpt does not provide training results or evidence of deployed on-device coaching.

### Source excerpt

⚠ Note: this post is AI-generated. The text below was written by Claude, and documents findings from a series of Claude Code sessions working on this project -- the experiments, bugs and measurements described are ones that came out of those sessions. The engineering work is real and the numbers were measured rather than estimated, but the write-up is the model's own account of what it did, not a human's independent retelling of it.

## Scanning bus stop codes with ML Kit and Vision in the GalwayBus Compose Multiplatform app

DevFeed: [Scanning bus stop codes with ML Kit and Vision in the GalwayBus Compose Multiplatform app](<https://devfeed.tech/articles/scanning-bus-stop-codes-with-ml-kit-and-vision-in-the-galwaybus-compose-multiplatform-app-25195.md>)

Original publisher: [Read original article](<https://johnoreilly.dev/posts/galwaybus-scan-stop-kmp/>)

Published: 2026-08-06T23:00:00Z

Content type: article

Language: en

Sources: [John O'Reilly](<https://devfeed.tech/sources/john-o-reilly.md>)

Topics: [compose-multiplatform](<https://devfeed.tech/topics/compose-multiplatform.md>), [ML Kit](<https://devfeed.tech/topics/ml-kit.md>), [cameraX](<https://devfeed.tech/topics/camerax.md>), [User Interfaces](<https://devfeed.tech/topics/user-interfaces.md>), [Kotlin](<https://devfeed.tech/topics/kotlin.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [android](<https://devfeed.tech/tags/android.md>), [app](<https://devfeed.tech/tags/app.md>), [apple](<https://devfeed.tech/tags/apple.md>), [bus](<https://devfeed.tech/tags/bus.md>), [camera](<https://devfeed.tech/tags/camera.md>), [compose](<https://devfeed.tech/tags/compose.md>), [compose-multiplatform](<https://devfeed.tech/tags/compose-multiplatform.md>), [kmp](<https://devfeed.tech/tags/kmp.md>), [ml-kit](<https://devfeed.tech/tags/ml-kit.md>), [ocr](<https://devfeed.tech/tags/ocr.md>), [on-device](<https://devfeed.tech/tags/on-device.md>), [plate](<https://devfeed.tech/tags/plate.md>), [recognition](<https://devfeed.tech/tags/recognition.md>), [scan](<https://devfeed.tech/tags/scan.md>), [scanning](<https://devfeed.tech/tags/scanning.md>), [text](<https://devfeed.tech/tags/text.md>), [vision](<https://devfeed.tech/tags/vision.md>)

### AI overview

GalwayBus adds a Scan tab that reads six-digit bus stop codes and opens the matching stop's departures. Camera preview, OCR results handling, matching, and UI are coordinated in shared Compose Multiplatform code; Android uses CameraX with on-device ML Kit, while iOS uses AVCaptureSession with Apple Vision through Kotlin/Native bindings. Both platforms drop incoming frames while recognition is in progress instead of queuing them. Android camera permission is bridged through an Activity launcher and a suspend function callable by shared code. Vision's language correction is disabled to avoid turning digit sequences into words. The excerpt says matching extracts full six-digit runs and requires an exact stop-code match; it does not include the remainder of the matching implementation.

### Source excerpt

Every bus stop in Galway has a plate with a 6-digit stop code printed on it. We recently added a "Scan" tab to the GalwayBus app that lets you point the camera at that plate and jump straight to the stop's departures. The text recognition runs entirely on device (ML Kit on Android, Apple's Vision framework on iOS), with the camera preview, matching logic and UI all living in the shared Compose Multiplatform code.

## Adding a Koog AI assistant (with structured output) to the FantasyPremierLeague CMP sample

DevFeed: [Adding a Koog AI assistant (with structured output) to the FantasyPremierLeague CMP sample](<https://devfeed.tech/articles/adding-a-koog-ai-assistant-with-structured-output-to-the-fantasypremierleague-cmp-sample-25194.md>)

Original publisher: [Read original article](<https://johnoreilly.dev/posts/fpl-koog-structured-output/>)

Published: 2026-07-27T23:00:00Z

Content type: tutorial

Language: en

Sources: [John O'Reilly](<https://devfeed.tech/sources/john-o-reilly.md>)

Topics: [AI Chat](<https://devfeed.tech/topics/ai-chat.md>), [compose-multiplatform](<https://devfeed.tech/topics/compose-multiplatform.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [chat](<https://devfeed.tech/tags/chat.md>), [compose](<https://devfeed.tech/tags/compose.md>), [compose-multiplatform](<https://devfeed.tech/tags/compose-multiplatform.md>), [framework](<https://devfeed.tech/tags/framework.md>), [gemini](<https://devfeed.tech/tags/gemini.md>), [jetbrains](<https://devfeed.tech/tags/jetbrains.md>), [kmp](<https://devfeed.tech/tags/kmp.md>), [koog](<https://devfeed.tech/tags/koog.md>), [kotlin](<https://devfeed.tech/tags/kotlin.md>), [sample](<https://devfeed.tech/tags/sample.md>), [structured-output](<https://devfeed.tech/tags/structured-output.md>)

### AI overview

The FPL Assistant uses Koog structured output to return a short Markdown response plus exact player and fixture IDs. The app can resolve those IDs against its own data and render existing player and fixture cards, avoiding error-prone name matching. Its Koog tools expose player, fixture, and tracked mini-league data from the app. The excerpt also says the agent registers these tools, streams intermediate tool calls and errors to the UI, and uses a Gemini model with its executor on all platforms, which can be customized. The supplied text ends during the explanation of the agent's loops, so it does not establish the full implementation.

### Source excerpt

Koog is JetBrains' Kotlin framework for building AI agents. In this article we're going to use it to add an in-app "FPL Assistant" to the FantasyPremierLeague Compose Multiplatform sample. This is a chat screen where you can ask about players, fixtures and mini-league standings. The answers render as player and fixture cards rather than just as text.

## Adding embeddings/RAG support to the Koog-based AI agent in Confetti

DevFeed: [Adding embeddings/RAG support to the Koog-based AI agent in Confetti](<https://devfeed.tech/articles/adding-embeddings-rag-support-to-the-koog-based-ai-agent-in-confetti-25191.md>)

Original publisher: [Read original article](<https://johnoreilly.dev/posts/confetti-koog-rag/>)

Published: 2026-05-25T23:00:00Z

Content type: article

Language: en

Sources: [John O'Reilly](<https://devfeed.tech/sources/john-o-reilly.md>)

Topics: [Embeddings](<https://devfeed.tech/topics/embeddings.md>), [Retrieval Augmented Generation (RAG)](<https://devfeed.tech/topics/retrieval-augmented-generation-rag.md>), [AI Agent](<https://devfeed.tech/topics/ai-agent.md>), [AI Chat](<https://devfeed.tech/topics/ai-chat.md>), [GraphQL](<https://devfeed.tech/topics/graphql.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agent](<https://devfeed.tech/tags/ai-agent.md>), [aiagent](<https://devfeed.tech/tags/aiagent.md>), [api](<https://devfeed.tech/tags/api.md>), [apollo](<https://devfeed.tech/tags/apollo.md>), [compose](<https://devfeed.tech/tags/compose.md>), [compose-multiplatform](<https://devfeed.tech/tags/compose-multiplatform.md>), [conference](<https://devfeed.tech/tags/conference.md>), [embeddings](<https://devfeed.tech/tags/embeddings.md>), [graphql](<https://devfeed.tech/tags/graphql.md>), [implementation](<https://devfeed.tech/tags/implementation.md>), [kmp](<https://devfeed.tech/tags/kmp.md>), [koog](<https://devfeed.tech/tags/koog.md>), [kotlin](<https://devfeed.tech/tags/kotlin.md>), [llm](<https://devfeed.tech/tags/llm.md>), [multiplatform](<https://devfeed.tech/tags/multiplatform.md>), [rag](<https://devfeed.tech/tags/rag.md>), [recommendation](<https://devfeed.tech/tags/recommendation.md>), [sample](<https://devfeed.tech/tags/sample.md>), [speaker](<https://devfeed.tech/tags/speaker.md>), [tools](<https://devfeed.tech/tags/tools.md>)

### AI overview

The excerpt explains how semantic retrieval can help Confetti's Koog-based agent find conference sessions for topic queries such as "what AI talks are on?" even when session titles use different wording. It describes embedding session text and user queries as vectors, comparing them by cosine similarity, and returning the closest sessions through an agent tool as RAG context. The article says the implementation adds Koog embeddings and RAG modules, vector storage, and Okio for a multiplatform on-disk embedding cache; it uses Gemini for both the agent's LLM and the embedding model. The supplied text cuts off during implementation, so it does not establish further setup details or results.

### Source excerpt

Recently I started looking at reworking the recommendation feature in the Confetti Compose Multiplatform sample (which uses Apollo Kotlin to consume a GraphQL API serving data for various conferences). The new implementation uses a Koog AIAgent with a ToolRegistry that includes session and speaker tools, giving the LLM on-demand access to the conference data it needs.

## Using Navigation 3 with Compose Multiplatform

DevFeed: [Using Navigation 3 with Compose Multiplatform](<https://devfeed.tech/articles/using-navigation-3-with-compose-multiplatform-25199.md>)

Original publisher: [Read original article](<https://johnoreilly.dev/posts/navigation3-cmp/>)

Published: 2025-11-15T00:00:00Z

Content type: tutorial

Language: en

Sources: [John O'Reilly](<https://devfeed.tech/sources/john-o-reilly.md>)

Topics: [compose-multiplatform](<https://devfeed.tech/topics/compose-multiplatform.md>), [Jetpack Compose](<https://devfeed.tech/topics/jetpack-compose.md>), [jetpack-navigation](<https://devfeed.tech/topics/jetpack-navigation.md>), [navigation](<https://devfeed.tech/topics/navigation.md>), [User Interfaces](<https://devfeed.tech/topics/user-interfaces.md>)

Tags: [adaptive](<https://devfeed.tech/tags/adaptive.md>), [android](<https://devfeed.tech/tags/android.md>), [compose](<https://devfeed.tech/tags/compose.md>), [compose-for-desktop](<https://devfeed.tech/tags/compose-for-desktop.md>), [compose-multiplatform](<https://devfeed.tech/tags/compose-multiplatform.md>), [dependencies](<https://devfeed.tech/tags/dependencies.md>), [gradle](<https://devfeed.tech/tags/gradle.md>), [jetbrains](<https://devfeed.tech/tags/jetbrains.md>), [jetpack](<https://devfeed.tech/tags/jetpack.md>), [keys](<https://devfeed.tech/tags/keys.md>), [layout](<https://devfeed.tech/tags/layout.md>), [list](<https://devfeed.tech/tags/list.md>), [manage](<https://devfeed.tech/tags/manage.md>), [mapping](<https://devfeed.tech/tags/mapping.md>), [material3](<https://devfeed.tech/tags/material3.md>), [multiplatform](<https://devfeed.tech/tags/multiplatform.md>), [navigation](<https://devfeed.tech/tags/navigation.md>), [project](<https://devfeed.tech/tags/project.md>), [sample](<https://devfeed.tech/tags/sample.md>), [stack](<https://devfeed.tech/tags/stack.md>)

### AI overview

The article shows Navigation 3 in the shared code of a Compose Multiplatform sample, using JetBrains-maintained versions of the library. Navigation is managed through a back stack: app keys map to content, and destinations are opened or removed by adding or removing entries. The setup includes Navigation 3 core UI and Material3 Adaptive Navigation3 dependencies, plus top-level routes. For an adaptive list/detail layout, the example creates a navigation strategy and passes it to the relevant component; the layout adjusts as the window size changes. The excerpt says this works on Desktop, Android, and iOS, and points to the Navigation 3 Code recipes repository for other use cases.

### Source excerpt

Navigation 3 is a new Compose based navigation library from Google that's designed with ease of use and flexibility in mind.

## Deploying a Kotlin-based remote MCP Server to Google Cloud Run

DevFeed: [Deploying a Kotlin-based remote MCP Server to Google Cloud Run](<https://devfeed.tech/articles/deploying-a-kotlin-based-remote-mcp-server-to-google-cloud-run-25200.md>)

Original publisher: [Read original article](<https://johnoreilly.dev/posts/remote-mcp/>)

Published: 2025-07-26T23:00:00Z

Content type: tutorial

Language: en

Sources: [John O'Reilly](<https://devfeed.tech/sources/john-o-reilly.md>)

Topics: [Cloud Run](<https://devfeed.tech/topics/cloud-run.md>), [Model Context Protocol](<https://devfeed.tech/topics/model-context-protocol.md>), [Google Cloud Platform (GCP)](<https://devfeed.tech/topics/google-cloud.md>)

Tags: [based](<https://devfeed.tech/tags/based.md>), [claude](<https://devfeed.tech/tags/claude.md>), [cloud](<https://devfeed.tech/tags/cloud.md>), [cloud-run](<https://devfeed.tech/tags/cloud-run.md>), [deployment](<https://devfeed.tech/tags/deployment.md>), [desktop](<https://devfeed.tech/tags/desktop.md>), [google](<https://devfeed.tech/tags/google.md>), [google-cloud](<https://devfeed.tech/tags/google-cloud.md>), [google-cloud-run](<https://devfeed.tech/tags/google-cloud-run.md>), [kotlin](<https://devfeed.tech/tags/kotlin.md>), [local](<https://devfeed.tech/tags/local.md>), [mcp](<https://devfeed.tech/tags/mcp.md>), [mcp-server](<https://devfeed.tech/tags/mcp-server.md>), [mobile](<https://devfeed.tech/tags/mobile.md>), [protocol](<https://devfeed.tech/tags/protocol.md>), [remote](<https://devfeed.tech/tags/remote.md>), [remote-mcp-server](<https://devfeed.tech/tags/remote-mcp-server.md>), [sdk](<https://devfeed.tech/tags/sdk.md>), [setup](<https://devfeed.tech/tags/setup.md>), [support](<https://devfeed.tech/tags/support.md>)

### AI overview

The ClimateTraceKMP sample's Kotlin MCP server can be deployed to Google Cloud Run by creating and selecting a Google Cloud project, building and publishing a container image to Artifact Registry with the Jib Gradle Plugin, then deploying the image. The server uses a Kotlin MCP SDK Ktor extension to provide SSE support. Once configured, the remote server can be accessed from Claude Desktop and mobile apps, and tested with MCP Inspector. The excerpt describes the setup but omits the actual configuration and command details.

### Source excerpt

I wrote a previous post about how to develop an MCP Server using the Kotlin MCP SDK. That was primarily based on local deployment (using stdio protocol)...a setup that Claude Desktop for example could only support at the time. With the announcement that Claude now supports access to remote MCP Servers (from desktop and mobile) I thought I'd take a look at deploying the MCP Server in the ClimateTraceKMP sample to Google Cloud Run.

## Using Google's Agent Development Kit for Java from Kotlin code

DevFeed: [Using Google's Agent Development Kit for Java from Kotlin code](<https://devfeed.tech/articles/using-google-s-agent-development-kit-for-java-from-kotlin-code-25197.md>)

Original publisher: [Read original article](<https://johnoreilly.dev/posts/kotlin-adk/>)

Published: 2025-07-10T23:00:00Z

Content type: tutorial

Language: en

Sources: [John O'Reilly](<https://devfeed.tech/sources/john-o-reilly.md>)

Topics: [AI Agent](<https://devfeed.tech/topics/ai-agent.md>), [Google](<https://devfeed.tech/topics/google.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [Kotlin](<https://devfeed.tech/topics/kotlin.md>), [Java](<https://devfeed.tech/topics/java.md>), [Model Context Protocol](<https://devfeed.tech/topics/model-context-protocol.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [ai](<https://devfeed.tech/tags/ai.md>), [development](<https://devfeed.tech/tags/development.md>), [google](<https://devfeed.tech/tags/google.md>), [java](<https://devfeed.tech/tags/java.md>), [kotlin](<https://devfeed.tech/tags/kotlin.md>), [kotlin-multiplatform](<https://devfeed.tech/tags/kotlin-multiplatform.md>), [mcp](<https://devfeed.tech/tags/mcp.md>), [mcp-server](<https://devfeed.tech/tags/mcp-server.md>)

### AI overview

A tutorial on using Google's Agent Development Kit for Java from Kotlin code, demonstrated in a Kotlin Multiplatform sample. It covers configuring Gradle dependencies, selecting LLM providers, creating and running agents, adding MCP and local function tools, handling Kotlin suspend functions, and using the ADK developer UI.

### Source excerpt

Google's Agent Development Kit for Java is described as "an open-source, code-first Java toolkit for building, evaluating, and deploying sophisticated AI agents with flexibility and control". In this article we're going to see how it can be consumed in Kotlin code (specifically in the agents module in the ClimateTrace Kotlin Multiplatform sample).

## Initial exploration of using Koog for developing Kotlin based AI agents

DevFeed: [Initial exploration of using Koog for developing Kotlin based AI agents](<https://devfeed.tech/articles/initial-exploration-of-using-koog-for-developing-kotlin-based-ai-agents-25198.md>)

Original publisher: [Read original article](<https://johnoreilly.dev/posts/kotlin-koog/>)

Published: 2025-06-21T23:00:00Z

Content type: tutorial

Language: en

Sources: [John O'Reilly](<https://devfeed.tech/sources/john-o-reilly.md>)

Topics: [Kotlin](<https://devfeed.tech/topics/kotlin.md>), [Model Context Protocol](<https://devfeed.tech/topics/model-context-protocol.md>), [IntelliJ IDEA](<https://devfeed.tech/topics/intellij-idea.md>), [SDK](<https://devfeed.tech/topics/sdk.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agent](<https://devfeed.tech/tags/ai-agent.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [android](<https://devfeed.tech/tags/android.md>), [android-studio](<https://devfeed.tech/tags/android-studio.md>), [compose](<https://devfeed.tech/tags/compose.md>), [compose-ui](<https://devfeed.tech/tags/compose-ui.md>), [custom](<https://devfeed.tech/tags/custom.md>), [framework](<https://devfeed.tech/tags/framework.md>), [kmp](<https://devfeed.tech/tags/kmp.md>), [koog](<https://devfeed.tech/tags/koog.md>), [kotlin](<https://devfeed.tech/tags/kotlin.md>), [kotlin-multiplatform](<https://devfeed.tech/tags/kotlin-multiplatform.md>), [kotlinconf](<https://devfeed.tech/tags/kotlinconf.md>), [mcp](<https://devfeed.tech/tags/mcp.md>), [mcp-servers](<https://devfeed.tech/tags/mcp-servers.md>), [multiplatform](<https://devfeed.tech/tags/multiplatform.md>), [sdk](<https://devfeed.tech/tags/sdk.md>)

### AI overview

The example uses Koog to build a single-run Kotlin agent with Google Gemini that requests climate emissions data, generates a Compose UI and preview, and adds the code to an Android Studio project. It uses configured MCP servers for climate data and IDE control. The excerpt says Koog also supports workflow agents, Google, OpenAI, Anthropic, OpenRouter, and Ollama providers, multi-provider prompt executors, and local annotation-based tools. It demonstrates switching between a local tool and an MCP-based tool; it does not provide comparative provider results.

### Source excerpt

Koog, announced recently at KotlinConf, is a new Kotlin-based framework designed to build and run AI agents. This article will outline initial exploration of using Koog along with a number of configured MCP servers (specifically mcp-jetbrains which we'll use to control the IntelliJ/Android Studio IDE and also one based on the ClimateTraceKMP Kotlin Multiplatform (KMP) sample built using the Kotlin MCP SDK).