Merchants today have to navigate across platforms like Swiggy, Zomato, Magicpin, and ONDC which leads to operational overload. Managing each one separately means constant logins, repeated updates, and missed opportunities — whether it's forgetting to run a promo or reacting too late to a rating drop.

Our web portal fixes this by becoming the one-stop assistant every merchant needs:

One place for everything: See orders, views, offers, and alerts across all platforms in a single dashboard.

We use Sarvam’s speech-to-text, merchants can speak in their native language to get things done.

Smart automation: With OpenAI’s LVMs, browser workflows like updating offers or running campaigns are handled automatically

Real-time alerts & insights: Get notified about issues or opportunities before they impact revenue.

And it's built to scale — lightweight, deployable via Kubernetes, and ready for thousands of merchants.

Building this in a tight 24-hour window came with its fair share of hurdles:

Agentic workflows were tricky: Setting up LVMs (like OpenAI’s browser agents) to reliably navigate dynamic, login-protected portals like MagicPin wasn’t straightforward. Handling session management, unpredictable DOM changes, and ensuring consistent results took multiple retries.

UI under pressure: Designing a clean, responsive UI that could display dynamic analytics, real-time alerts, and control workflows — all while keeping it simple for merchants — was a challenge. Balancing functionality with speed in limited time meant prioritizing what really mattered.

Speech-to-text integration: Plugging in Sarvam’s speech model to accept voice input in multiple languages required handling edge cases like background noise, mixed-language inputs, and making sure the UI responded instantly after transcription.

Security & session handling: Logging into third-party portals on behalf of merchants introduced concerns around token storage, credential safety, and session expiry — solving this securely within a hackathon window took creative shortcuts.

Debugging under load: With multiple async components — speech, vision models, browser agents — debugging workflows in real-time across different machines and browsers became a race against the clock.