
Weekend Creative Agent Challenge: DevRel Morning Drop
**DevRel Morning Drop** is an always-on AI agent built with **AWS Amplify Gen 2 and Qwen API** that automatically delivers fresh developer learning ideas, project prompts, content concepts, explanations, and technical challenges every morning.
Vision & What It Does
Every developer I know has had the same kind of morning: there is time to learn or build something, but the blank page wins. βWhat should I work on today?β becomes ten minutes of searching, scrolling, and reconstructing context. An AI chatbot can suggest an idea, but it still waits for me to open it, explain who I am, and ask the right question.
I built DevRel Morning Drop around a different idea: the best developer assistant should have already done the preparation.
Morning Drop is an always-on Developer Relations learning and creativity agent. A user signs in once and tells it the topics they care about, their experience level, their goals, the time they normally have, and the categories they want. From then on, an autonomous scheduled workflow prepares a fresh personalized pack before the user returns.
Each Drop contains up to six connected pieces: something to Learn, something small to Build, a concept to Explain, a piece of technical content to Create, a deliberate-practice Challenge, and a playful Wildcard. The sections are intentionally practical. A Build card might ask an intermediate serverless developer to make a tiny webhook queue with API Gateway, SQS, and Lambda. The Explain card can turn the same underlying idea into a cafΓ© analogy for a beginner audience. The Create card might turn it into a 30-second DevRel reel.
There is no Generate button. When the user opens the app, the headline is simply: βYour Drop is ready.β
The application also includes an Agent Status view. It shows the last autonomous run, the next scheduled creation, the outcome, and processing stages such as profile loading, recent-topic analysis, Qwen generation, validation, and persistence. Agent Activity stores genuine run records rather than pretending local sample content was autonomously generated. That visible evidence is important: it makes the always-on behavior part of the product experience instead of an invisible backend claim.
The Idea Library adds a complementary on-demand path for moments when a developer already has a direction. A signed-in user can click Ask Qwen, choose a category, describe the outcome, and receive one focused idea shaped by their saved profile. A protected Amplify custom mutation invokes a separate Lambda, validates the response, and saves the idea to that user's owner-isolated library. The autonomous Drop remains the main experience; this action simply makes the growing idea archive more useful.
Β
[SCREENSHOT: Todayβs Morning Drop]Β How I Built It
I started with the autonomous path rather than the visual dashboard. The central workflow is:
schedule β load profiles β check local date β read recent Drops β create run record β call Qwen β validate JSON β save Drop β complete run recordThe frontend is React and TypeScript with Vite, Tailwind CSS, Lucide icons, and a small amount of Framer Motion. The visual direction is closer to a developer notebook crossed with GitHub and Linear than a generic AI dashboard. It uses one lime accent, restrained borders, dense technical metadata, and no prompt box.
AWS Amplify Gen 2 defines the backend in TypeScript. Amplify Auth provides an Amazon Cognito email sign-up and recovery flow. Amplify Data defines three owner-isolated models: Β
UserProfileΒ , Β MorningDropΒ , and Β AgentRunΒ . The signed-in user can only access their own records. The scheduled Lambda has separate IAM-scoped access through Amplify's resource authorization.For generation I created an Β
AIProviderΒ interface and a Β QwenProviderΒ implementation. Qwen is called through ModelScope's OpenAI-compatible endpoint, but the request exists only inside the AWS Lambda function. The ModelScope token is referenced with Amplify Β secret()Β and is never sent to the browser, placed in a public environment variable, or committed to the repository.I did not trust free-form model output. The prompt requires a single JSON object with exact keys for the theme, opening message, and six sections. The streamed response is collected, the JSON object is extracted, and Zod validates required fields and useful length bounds. Only a valid object is stored. If generation or validation fails, the run becomes Β
FAILEDΒ , CloudWatch receives a concise diagnostic, and no partial Drop is written. Previous Drops remain available.Variety was another key decision. Before each generation, the function reads recent Drop titles and sends them back as Β
avoidTopicsΒ . This makes repetition a data problem instead of relying on a vague βbe originalβ instruction. The day of the week supplies a gentle rotating theme, but it never overrides the user's interests.Cost control shaped the architecture too. The worker checks whether a Drop already exists for that profile and local date. It generates all sections in one capped request, stores the result, and never calls Qwen while rendering the application.
On-demand requests are also explicit and capped. Opening or searching the Idea Library never invokes the model; only the signed-in Ask Qwen submission calls the dedicated function.
Β
[SCREENSHOT: Expanded Build card]Β AWS Services Used / Architecture Overview
The project uses:
- AWS Amplify Gen 2 Hosting for the fullstack branch deployment and React application
- Amplify Auth / Amazon Cognito for user identity
- Amplify Data / AWS AppSync / Amazon DynamoDB for the typed API and durable records
- Amplify Functions / AWS Lambda for autonomous generation
- Amazon EventBridge Scheduler for the daily invocation
- AWS Systems Manager Parameter Store through Amplify secrets for the ModelScope credential
- Amazon CloudWatch Logs for operational diagnostics
ββββββββββββββββββββββββββββ β AWS Amplify Hosting β β React + TypeScript β ββββββββββββββ¬ββββββββββββββ β Cognito + AppSync β βββββββββββββββββΌββββββββββββββββ β UserProfile / MorningDrop / β β AgentRun in DynamoDB β βββββββββββββββββ²ββββββββββββββββ β EventBridge Scheduler βββΊ Lambda Morning Agent βββΊ Qwen / ModelScope β Zod validation + CloudWatch loggingFor this weekend MVP, one universal schedule runs at 00:00 UTC. The function computes the current date in each user's stored IANA timezone, and the UI surfaces the Drop by local date. A future version can create grouped timezone schedules, but the universal window is predictable, inexpensive, and easy to explain.
Β
[SCREENSHOT: Agent Status and Agent Activity]Β Challenges and Key Decisions
The hardest product decision was resisting the temptation to add manual generation. That button would make demos easier, but it would weaken the entire challenge narrative. I kept testing as an administrator-only Lambda invocation and clearly label local historical sample data as demo seed content.
The second challenge was balancing personalization with completion time. Six unrelated ideas would feel like noise. The prompt therefore asks the sections to reinforce each other without duplicating one another and gives Qwen a single total time budget. The user's enabled categories control what the UI presents.
The third challenge was keeping autonomous access secure. Browser clients use Cognito user-pool authorization and owner rules. The scheduler cannot impersonate each user, so the Lambda receives its own narrowly scoped resource access and explicitly writes the user's owner identity into generated records.
What I Learned
This project changed how I think about agent UX. The most persuasive evidence of autonomy is not an βAIβ badge; it is a useful artifact with a timestamp and a trustworthy trail showing how it arrived. The Agent Activity page became as important as the generated cards.
I also learned how naturally Amplify Gen 2 can mix high-level primitives with deeper AWS services. Auth, Data, and Functions stay concise and typed, while Β
backend.tsΒ can add an EventBridge Scheduler resource with AWS CDK. That combination worked well for a weekend build: very little infrastructure glue, but no need to hide the real architecture.Finally, structured generation is worth the extra effort. A schema, validation, duplicate protection, and explicit failure state turn an interesting model response into a dependable product workflow.
Morning Drop is still an MVP, but the central promise works: I can leave, return, and find a developer idea that was made for me while I was away.
Try It / Source
- Live app: Β https://main.d2f9pifhpnibeu.amplifyapp.com/Β
- Public repository: Β https://github.com/BandhiyaHardik/Weekly_Challenge_DevRel_MorningΒ
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article