DEV Community: Yeshua

Building Reptrainer: The AI "Flight Simulator" for Sales with Gemini Live

Yeshua — Mon, 16 Mar 2026 16:21:30 +0000

Disclaimer: This piece of content was created for the purpose of entering the #GeminiLiveAgentChallenge hackathon.

Sales is a high-stakes performance, but until now, practicing for it has felt like a chore. Most representatives walk into critical discovery calls with high anxiety because they haven't "repped" the scenario in a realistic environment.

Enter Reptrainer (DealPilot): a real-time AI training ground built on the cutting edge of the Google Cloud AI ecosystem.

The Vision: A Safe Space for High-Pressure Conversations

Our goal was to create a "flight simulator" for sales. We wanted to move beyond the robotic, turn-taking chatbots of the past and build something that felt human. The release of the Gemini Live API was the missing link, finally allowing us to build a voice-first agent that is responsive, reactive, and capable of natural interruption.

The Tech Stack: Truly Multimodal

Building a system that listens, thinks, and speaks required a deeply integrated multimodal architecture that minimizes latency and maximizes realism.

1. The Voice Brain: Gemini Live API & Python ADK

The core interaction is powered by a Python backend using the Google Agent Development Kit (ADK) and FastAPI.

Bidirectional Streaming: We leverage WebSockets to stream audio directly to our ADK-powered agent.
Realistic Interruption (Barge-In): One of our proudest achievements was tuning Voice Activity Detection (VAD) to ensure the AI "stops" speaking instantly when the user speaks - just like a real prospect would.

2. The Unified Multimodal Debrief

Sales training is only as good as the feedback. Traditionally, generating a visual and narrated report would require stitching together multiple services. Reptrainer achieves this in a single "leap" using Gemini's native multimodal output:

Unified Generation: Instead of separate calls for text and images, we use a single API call to generate both coaching insights and contextually coherent infographics interleaved in the response.
Contextual Coherence: Because the model sees the coaching text and the image generation as part of the same task, the visuals directly and accurately reflect the specific feedback given to the rep.

3. Grounded Intelligence with Vertex AI

A "generic" buyer persona is useless for enterprise sales. We use Vertex AI grounding to ensure our personas are aware of your actual product documentation and competitor "battle cards." If you mention a specific feature or rival, the AI knows exactly how to push back based on your real-world market position.

What We Learned (The Hard Way)

Building with Gemini Live taught us several key lessons about the future of voice AI:

Brevity is King: In voice interactions, long responses feel like monologues. We learned to implement strict brevity guards to keep the flow conversational.
Managing "Dead Air": Executing tools (like real-time data lookups) takes time. We coached our AI to use natural filler phrases like "Hang on, let me check that pricing for you..." to maintain immersion while the tools execute in the background.

Infrastructure: Scalability with Google Cloud

The entire system is hosted on Google Cloud Run, orchestrated by Cloud Build for a seamless CI/CD pipeline. We use Firebase for real-time state synchronization, Firestore for session persistence, and Cloud Storage for hosting multimodal assets.

The unified Google Cloud ecosystem allowed us to build features in days that would have normally taken weeks of "glue code" with disparate providers.

What’s Next?

We’re just scratching the surface. Our roadmap includes integrating Gemini Vision to provide real-time coaching on body language, and simulating "Buying Committees" where you pitch to multiple AI stakeholders at once.

Building Reptrainer has shown us that the future of sales training isn't just AI—it's Live AI.

Created for the #GeminiLiveAgentChallenge hackathon.

I created a Christian "Bible Gym" webapp

Yeshua — Mon, 02 Mar 2026 03:32:04 +0000

This is a submission for the DEV Weekend Challenge: Community

The Community

I built this for my fellow Christians.

You know how it is. We spend hours at the gym tracking our macros, our hit-rate on the court, or our daily steps on our watches. But when it comes to our spiritual life, most of us (myself included!) are basically "couch potatoes." We'd read a verse here and there, but nothing would really stick. No muscle was being built.

I wanted to bridge that gap. This web app is for the Christian believers who want to take their Scripture study from a "passive hobby" to a "high-intensity workout." It’s for the church small groups, the youth ministries, and the personal prayer warriors who want to be "fit" for every good work as Apostle Paul says in 2 Timothy 3:17 (KJV)

What I Built

I built Bible Gym - a gamified, high-energy web app designed to bake Scripture into your daily routine using proven athletic training metaphors.

Instead of just reading, you're training. Every day, the app generates a "Daily Workout" consisting of 4 intense drills. Here are some of the features Bible Gym that make it an awesome and helpful tool for Christians:

AI-Powered Daily Workout: Always-fresh batch of daily drills generated with Gemini. If Gemini fails because of quota or any other reason, I use the bible-api as backup.
Memorization Drill: Clozed-deletion (filling in the blanks) to build active recall.
Context Challenge: Rapid-fire questions to ensure you're using "proper form" (understanding the who, what, and why of a passage).
Verse Match: High-speed matching to build quick-twitch Scripture recognition.
Rearrange Drill: Ordering verses to strengthen your grasp of the Bible's structure.
Mastery Mode: Where you push specific verses through 5 levels of difficulty until they are etched into your heart forever.
Training Groups: Join a group or create one yourself; because everyone trains better with a team.

Demo

Live App

The app features a "Neo-Brutalist" design with vibrant colors, thick borders, and high-energy animations. It feels more like a fitness app than a dusty library.

Code

Check out the full monorepo here:

lucky-chap / bible-gym

🕊️ Bible Gym — Train Your Spirit

"Physical training is good, but training for godliness is much better, promising benefits in this life and the life to come." — 1 Timothy 4:8

Welcome to Bible Gym, a high-intensity developer-led experiment in spiritual fitness. We took the psychology of a modern fitness app—the streaks, the stats, the "just one more set" addiction—and applied it to the most important "muscle" you own: your spirit.

💪 The Workout Philosophy

Bible Gym isn't a library; it's a training ground. We believe that Scripture should be etched into the heart, not just glanced at on a screen. Every day, the gym generates a Global Daily Workout for the entire community.

🏋️ The Drills

Memorization Drill: Clozed-deletion recall. Fill in the missing words to build active retrieval pathways.
Context Challenge: Rapid-fire questions on authorship, history, and theme. Proper form means understanding the…

View on GitHub

How I Built It

This was an absolute blast to build. Here’s the stack under the hood:

Next.js (App Router): For that smooth, server-side rendered performance.
Appwrite: My backend powerhouse. It handles all the user authentication, real-time database syncing for scores/streaks, and local-to-cloud persistence.
Framer Motion: For those "punchy" micro-animations that make the UI feel alive and responsive.
Lucide Icons: Keeping the "Gym" aesthetic consistent across every drill.
Gemini API: I integrated AI to dynamically generate context questions and themed workouts based on whatever topic the community is feeling that day.
Vercel Cron Job: An automated job to handle the daily workout flip at midnight UTC, ensuring there's always a fresh batch of verses waiting for the believer every morning.

I pray you find peace, internet friend. 🕊️💪

Typecraft: Practice typing in a new way

Yeshua — Mon, 27 Jan 2025 05:50:04 +0000

This is a submission for the Agent.ai Challenge: Full-Stack Agent and Productivity-Pro Agent (See Details)

What I Built

I built TypeCraft.
Typing games have long been a staple in the world of educational and entertainment software, helping users sharpen their typing skills while immersing them in enjoyable challenges. But what if a typing game could go beyond generic scenarios and offer uniquely crafted experiences tailored to each player’s input?

The Concept: Typing Meets AI-driven Customization

Imagine a typing game where the storyline evolves based on the player’s choices. For instance, a player might input a specific topic, character type, or difficulty level, and the game dynamically generates a scenario that weaves these elements into the gameplay. This customization not only enhances user engagement but also fosters creativity and provides endless replayability.

Potential Use Cases

Education: Schools can use the game to teach typing in an engaging way, allowing students to explore historical events, scientific concepts, or literary genres through their own chosen scenarios.
Entertainment: Typing enthusiasts seeking a unique experience will enjoy the endless possibilities offered by AI-driven content generation.
Professional Training: Organizations can customize the game to simulate real-world typing tasks, such as drafting emails or coding.

Demo

Try out the agent over here: TypeCraft

Github Repo

Client
API

Agent.ai Experience

While Agent.ai offers powerful tools for rapid prototyping and dynamic content generation, working with the platform wasn’t without its challenges. There were times when the API didn’t run as expected, or POST requests would fail to send consistently, causing interruptions in development. These moments of frustration highlighted areas where the platform could improve its reliability and robustness.

Despite these setbacks, Agent.ai proved to be an invaluable asset. Its straightforward user interface made it easy to prototype ideas quickly, allowing for rapid iteration and testing. The platform’s ability to generate creative and immersive scenarios showcased its potential for building innovative applications. For developers willing to navigate occasional hiccups, Agent.ai remains a strong choice for bringing AI-driven ideas to life.

Bringing people together with Neighborly

Yeshua — Mon, 20 Jan 2025 01:16:29 +0000

This is a submission for the GitHub Copilot Challenge : Fresh Starts

What I Built

I built Neighborly, a platform designed to bring local communities and neighborhoods together. By fostering connections and encouraging participation, we help you make informed decisions and foster a sense of unity and growth, building a stronger, more united community.

With Neighborly, community leaders can:

Create an online community for their local community
Send invites: Seamlessly invite new members to join the platform.
Gather feedback: Enable members to share their thoughts and ideas to improve the community.

Neighborly’s mission is to strengthen relationships, encourage collaboration, and create more engaged and active local communities.

How was this built in 24 hours?

The simple answer is planning. Within 3 days I had settled on what type of app to build based on what was going on around me. I then used the rest of the days to search out for templates and components to use. Truth is, I had previously written similar logic implemented in Neighborly about a month ago or so over here, so integrating it wasn't so hard for me. The main challenge was seeing the end of the app from the beginning. This is what I mean: I could envision the app before I built it, but nothing always goes as planned. I had to keep refining ideas to really build out what I saw in my mind. Time was also not really on my side so MVP was the only reasonable thing to do. Here are some of the resources I used:

Simplified App Anatomy

A member can't be a part of multiple communities (Since you can't be a member of two physical communities at the same time)
If you are a member of a community, you can't join another community unless your community leader removes you from your current community.

Demo

Try out the app here: https://neighborly-github.vercel.app

Repo

https://github.com/lucky-chap/neighborly

Copilot Experience

GitHub Copilot was an invaluable assistant throughout the development of Neighborly. Here are some of the ways it helped:

Idea Generation

When brainstorming features for the MVP, I used Copilot to prototype several ideas quickly. By providing simple prompts like “What main features does this app need?” or “Which features define the core of the app?” I was able to iterate rapidly.

Code Suggestions and Autocomplete

Copilot’s autocomplete feature was particularly helpful when writing repetitive or boilerplate code, such as form validation, component generation and API integration with Convex. It learned from my convex files and could further help me by providing boilerplate. Another thing is, it assisted in creating a REST API endpoint to handle file upload with minimal manual effort. All I had to do was tweak it a little bit to fit what was in mind.

Error Resolution

Debugging became more efficient with Copilot’s suggestions. It not only pointed out syntax errors but also recommended fixes for common pitfalls, like explicitly infering types on objects. Since Convex uses TypeScript, it was also way easier for Copilot to provide suggestions based on the type of data coming from the Convex backend.

GitHub Copilot Chat

The chat feature was a game-changer for seemingly easy yet boring and time-consuming tasks. I used it to change parts of my code that would have taken too long for me to manually change or update.

Model Switcher

Experimenting with different GitHub models allowed me to tailor the code generation to the app’s specific needs. The models provided a balance of creativity and precision that made development smoother. The variety of outputs was something I thought was cool. (I think Claude is more accurate)

GitHub Models

I unfortunately could not use GitHub Models this time. I hope to use it some other time in my own projects and see what it's all about.

Conclusion

Participating in the GitHub Copilot 1-Day Build Challenge was an enriching experience. Copilot’s capabilities allowed me to focus more on the creative aspects of building Neighborly, while minimizing the time spent on repetitive and boring tasks. The result is an MVP that showcases the potential for technology to bring communities closer together.

Looking ahead, I’m excited to expand Neighborly with more features, such as direct messaging and advanced event management. I also plan to integrate user feedback into the development process to ensure that Neighborly meets the evolving needs of its communities.

Thank you, GitHub Copilot, for making the development process seamless and enjoyable!

Next.js, Tailwind CSS and Shadcn-ui template

Yeshua — Sun, 06 Aug 2023 11:15:42 +0000

In my previous post, I introduced Kaminari, a Next.js and Tailwind CSS template.

I have since updated it to use shadcn-ui and now has dark mode.

Here are a few demo images:

It is configured with:

Next.js 13 (with the new app directory 🤯)
Tailwind CSS 3 ✨
shadcn-ui
Dark mode 🌒
TypeScript 👍
Prettier 🦋
Eslint 👌

Make sure to check out the below links and drop a star if you do not mind!

GitHub Repo
Live Demo

Next.js Boilerplate

Yeshua — Sun, 27 Nov 2022 09:07:24 +0000

Introducing Kaminari, a Next.js and TailwindCSS boilerplate that gets you up to speed with new projects. No need to go through all the setups and configurations.

It is configured with:

Next.js 13 (with the new app directory 🤯)
TailwindCSS 3 ✨
TypeScript 👍
Prettier 🦋
Eslint 👌
Commitlint 🔗
LintStaged ➡️
Husky 🐶
Seo 🌐

This is a work in progress and I'd like to get feedback from the community. Please do well star and spread the word. Thanks ✨✅

GitHub Repo
Live Demo