CariMurah

Inspiration

In a rapidly shifting global economy, consumers are constantly searching for ways to stretch their hard-earned currency. Yet, price comparison remains a tedious maze of visiting countless ecommerce platforms, parsing mismatched catalog descriptions, and manually tracking fluctuating deals. This pain is magnified for bilingual communities who struggle with search tools optimized for only one language.

We built CariMurah ("Find Cheap" in Indonesian) to democratize smart target-shopping. But building a consumer application is only half the battle. Our second biggest inspiration was the modern "Build in Public" movement: we wanted to create an app that didn't just solve a consumer problem, but used cutting-edge automated product analytics (Novus.ai by Pendo) to instrument itself, spot onboarding bottlenecks, and streamline engineering workflows automatically.

What it does

CariMurah is a complete, production-ready mobile and web ecosystem that turns complex pricing searches into single-button actions.

• Bilingual Adaptability & Universal Scanning: Users can seamlessly toggle between English and Indonesian. The camera scanner allows them to upload or take a direct photo of any consumer item.
• Transparent AI Processing Pipeline: When scanning, the app visually guides the user with a real-time, step-by-step status engine (e.g., AI is gathering online prices, Contacting smart agents, Parsing local datasheet specs). This keeps the user engaged and builds trust during multi-second calculations.
• NLP Smart Database Assistant: Features a natural language conversational hub integrated with our high-performance MongoDB database, allowing users to query transaction logs, historical price limits, and active inventory in plain human language.
• Smart Watchlist & Live Thresholds: Users can add products to a custom Watchlist and configure exact target-price thresholds to trigger alerts when prices drop.
• B2B/Enterprise Analytics & Automated Reports: Generates structured monthly summaries, PDF-ready savings insights, and exportable CSV tables with a single tap.

How we built it

To deliver a robust, beautiful experience on both mobile and web, we integrated a modern full-stack architecture:

• Frontend & Visuals: Built a fully responsive frontend utilizing React 18, Vite, Tailwind CSS, and Motion for smooth screen-to-screen transitions. Icons are powered dynamically by lucide-react. The app is compiled and fully wrapped to be friendly for direct Google Play Store mobile deployment.
• Back-End & Core AI: Engineered a secure Node/Express server acting as an API gateway. The price search leverages Google Cloud Agent Builder (Discovery Engine Datastores and Dialogflow CX) mapped side-by-side with localized data inside MongoDB.
• Automatic Instrumentation via Novus.ai (by Pendo): Instead of spending hours manually configuring telemetry script rules or coding analytic event trackers, we integrated Novus.ai. Novus instantly mapped our repository structure, auto-instrumented critical user checkout and onboarding funnels, and tracked active user retention. If an analytic script error or configuration mismatch pops up, Novus immediately spots it and can suggest direct pull requests right to our repository.

Challenges we ran into

• UI Layout Bilingual Consistency: Indonesian text length differs dramatically from English. We spent multiple iterations perfecting our typography and padding rules to ensure text containers and progress bars never clipped or broke the responsive visual rhythm.
• GCP IAM & Discovery Engine Permissions: Aligning access controls for the live Datastore endpoints during quick iterations caused permission denials initially, which we resolved by optimizing service account roles.
• Real-time Agent Telemetry: Integrating standard Pendo/Novus event hooks in a sandbox preview environment while keeping local tests building green required carefully crafting type safety definitions within global.d.ts and pendo.d.ts.

Accomplishments that we're proud of

• Play Store Compilation Ready: We completed a fully optimized client build that is ready for deployment, allowing users to download and try CariMurah directly on their native screens.
• Zero-Overhead Analytics Setup: By relying on Pendo Novus, we achieved full product-funnel instrumentation without writing a single line of tedious tracker boilerplates.
• High Visual Polish: Crafted a distinctive, beautiful display interface utilizing spacious off-white and elegant navy slate themes, moving entirely away from default templates.

What we learned

• The Power of Build-In-Public Telemetry: Integrating with Pendo’s latest product agent changed how we think about deployments. We learned that software doesn't just get built—it instruments itself and guides the developers automatically.
• Asymmetric API Concurrency: We learned how to parallelize heavy Google Cloud Agent Builder queries alongside local MongoDB transactions, keeping response lag below the user-perceived threshold.

What's next for CariMurah

• Cross-border Shipping Integrations: Connect automated courier rates to give users the "True Total Cost" of imported goods.
• Predictive AI Price Modeling: Use historical pricing arrays in MongoDB to train localized regression models, advising users when exactly is the best week to buy an item.
• Global Merchant Marketplace: Allow local mom-and-pop shops to index their inventory directly into our datastore, driving traffic back to local physical retail stores.

BuildInPublic

Building CariMurah has been a transformative experience that completely reshaped our development workflow, engineering mindset, and overall productivity through two key strategies: implementing automated product analytics via Novus.ai and embracing a rigid "Build in Public" culture.

1. Devops & Workflow Automation (with Novus.ai) Implementing Novus.ai directly into our repository has pioneered an "autonomous diagnostic loop" that drastically reduces our Mean Time to Resolution (MTTR) and boosts productivity: Instant Automated Pull Requests: Instead of manually hunting down runtime dependency exceptions, API authentication issues, or translation boundary edge-cases, Novus tracking automatically identifies errors in our script flow and acts proactively by creating targeted Github Pull Requests to fix them. Onboarding & Retention Optimization: By tracking real-time onboarding rates and telemetry triggers (like scanning bilingual audio or initiating database inquiries), our product team gets immediate visual feedback on where users drop off, enabling instant, data-driven modifications. Streamlined Telemetry over Manual Logging: Novus offloads the need for manual, custom log management. It connects directly with our repository, giving us structured insights that connect issues back to precise lines of code.
2. Engineering Discipline & The "Build in Public" Feedbacks We committed to a transparent, high-integrity shipping workflow by releasing our app directly to the Google Play Store and documenting our journey online. Our LinkedIn post shows our dedication to this open loop:
3. LinkedIn Build-in-Public Showcase: View Our Public Journey & Launch Update
4. Real-world Feedback loops: Running the live mobile application on personal devices allowed us to get immediate, real-world user responses. It forced us to think deeply about localized, bilingual user interfaces and real-time step visualization (e.g., explaining API query stages while querying databases or calling GCP Agent Builder) to build trust during loading states.
5. True Full-stack Mastery: Building CariMurah advanced our mastery of complex asynchronous processing—handling heavy integrations such as GCP Discovery Engine/Agent Builder, MongoDB cloud data persistence, automated watchlist thresholds, and automated B2B reports generation within a single, elegant platform.

Built With

• gemini
• typescript