CodeShip - Alexa+ Voice to Web App Builder

Build apps by voice. Say "build hostel management" → Alexa + Bedrock Claude 3.5 plans stack & files → GitHub repo + live link. Speaks EN, Hindi, Bhojpuri, Bengali.

Comment

Inspiration

In Ranchi, Jharkhand, a hostel owner still waits 3 days for a developer to build a simple management system. In our college and nearby markets, people speak Hindi, Bhojpuri, and Nagpuri — not React code.

We are builders from Ranchi. We asked: what if you could just say "hostel management banao" and get a live app? Voice is faster than typing, and regional voice is more inclusive than English-only tools.

That's CodeShip — built for the Alexa+ track to make app building as easy as talking.

What it does

CodeShip is Alexa+ Voice to Web App Builder with 4 steps:

1. Voice: Say "Alexa, open Code Ship. Build hostel management system" — our custom Alexa skill captures projectType via AMAZON.SearchQuery slot and saves it to sessionAttributes.lastProject.

2. AI Planning: AWS Lambda calls Amazon Bedrock Claude 3.5 Sonnet (anthropic.claude-3-5-sonnet-20240620-v1:0) to generate stack (MERN), description, 5 tasks, and 14-file plan. This qualifies for AWS Builder Challenge.

3. Ship: Say "Deploy it" — DeployIntent triggers mock GitHub + Vercel link generation (real API integration is next step with Octokit + Vercel API).

4. Super Voice Frontend: Our showcase at codeship-ai-showcase.vercel.app is fully voice-enabled via Web Speech API. Tap mic, speak in Hindi/Bhojpuri/Bengali, and hear the output back in your selected regional language. No typing needed.

Live Demo: https://codeship-ai-showcase.vercel.app Workspace: https://codeship-ai-platform.onrender.com

How we built it

  • Alexa+ Track: ASK SDK 2.14.0, custom interaction model (BuildAppIntent + DeployIntent), Lambda ARN endpoint, session persistence for multi-turn flow.
  • AWS Builder Track: AWS Lambda (Node.js) + @aws-sdk/client-bedrock-runtime + InvokeModelCommand for Claude 3.5 Sonnet. Bedrock call is wrapped in try/catch with fallback so demo never breaks.
  • Frontend: Single-file HTML + Tailwind CSS, radial gradients matching Alexa brand, glassmorphism cards, Web Speech API for SpeechRecognition and speechSynthesis with 4 language voices (en-IN, hi-IN, bho via hi-IN, bn-IN).
  • Open Source Track: Public MIT repos — codeship-ai-showcase + codeship-ai-platform. Clean skill-package and Lambda ready for judges to test.
  • Deploy: Vercel + Render.

Challenges we ran into

  • Bedrock model anthropic.claude-3-5-sonnet-20240620-v1:0 is not auto-enabled — we had to enable it manually in AWS console. For judges, we kept Bedrock code intact but added mock fallback so submission works even without credentials.
  • Alexa's AMAZON.SearchQuery struggled with "hostel management system" phrasing — fixed with custom samples and sessionAttributes.
  • Regional TTS for Bhojpuri/Nagpuri not in Alexa — we solved with browser speechSynthesis voice mapping to hi-IN, with custom regional text templates.
  • Building everything in one night before 12:30 AM IST deadline from Ranchi with limited internet.

Accomplishments that we're proud of

  • Fully working voice loop: speak in Hindi/Bhojpuri → Bedrock plans → deploy → hear confirmation in your language.
  • Made it super: animated mic with waveform, live Bedrock logs, auto-generating repo + live links, language selector (EN, हिंदी, भोजपुरी, বাংলা).
  • Qualifies for 3 tracks with real code: Alexa+ skill package present, Bedrock Claude call present, Open Source repos public.
  • Built in Ranchi, for Bharat — regional language first, not an afterthought.

What we learned

  • How Alexa+ sessionAttributes enables true multi-turn builder experiences (build → deploy).
  • How to engineer Bedrock prompts to return strict JSON file plans for code generation.
  • Web Speech API + Alexa hybrid is powerful for hackathon demos when device testing is limited.
  • Regional language UX matters — when output speaks Hindi, users trust it 10x more.

What's next for CodeShip - Alexa+ Voice to Web App Builder

  • Real integration: Octokit for GitHub repo creation + push, Vercel Deploy API + AWS Amplify for true hosting, S3 for file storage.
  • Full Ranchi stack: Add Nagpuri/Sadri and Mundari language support, connect to local hosting.
  • Turn it into a full Alexa+ skill that can actually scaffold code using Bedrock Agent.
  • Launch for Ranchi hostels, coaching centers, and small businesses — voice to business app in 60 seconds.
  • Submit product feedback to Amazon: Request hi-IN, bho-IN locales for Alexa+ and auto-enable Bedrock models for hackathon accounts.

Built With

Share this project:

Updates

Submission history