Three products.

NovaX AI is the product I’ve spent the most time on. The first version was rough. The API routing was messy, the prompting was inconsistent, and the UX wasn’t right. I rebuilt it multiple times. Each rebuild was more focused than the last.

The current architecture is built around a FastAPI backend that handles LLM API calls, conversation state, and routing. The frontend is designed to stay out of the way — clean enough that users focus on the conversation, not the interface.

Saathi AI is built around simplicity. The conversational interface is intentionally straightforward — no complex navigation, no jargon, no overwhelming options. A user describes what they’re experiencing and gets clear, accessible guidance.

The product is deployed on Cloudflare Pages for maximum global performance. The AI layer is tuned specifically for health‑adjacent queries — conservative where it needs to be, informative where it can be, and always clear about what it can and cannot tell you.

This is the most technically ambitious product I’ve tried to build. The core challenge is orchestrating multiple data sources — hotel availability, transport options, local activity data, weather, budgeting — into coherent, personalized travel recommendations through a single conversational interface.

The architecture uses LLM agents that can reason about user preferences across multiple turns of conversation, remember context, and call external APIs to get real‑time data. The hard part isn’t any single capability — it’s making all of them work together reliably and feel natural in a conversation.