Com-Plan-ion

Inspiration

Event planning is broken in a way that everyone has experienced but nobody has fixed. The average corporate event planner spends 40+ hours on logistics — chasing vendor quotes over email, manually tracking budgets in spreadsheets, sending mass blast communications that feel impersonal, and still showing up on event day to surprises. We wanted to build something that felt less like software and more like having a brilliant, tireless co-planner who never forgets anything and never sleeps.

We were also excited by a specific challenge: could we make four very different sponsor technologies — an AI model, a document database, a blockchain, and a voice synthesis platform — feel like they were designed together rather than bolted on? That became the creative constraint that shaped every decision we made.

What We Learned

Building Com-Plan-ion taught us that the hardest part of integrating multiple AI systems isn't the individual APIs — it's making them feel like one coherent product. We learned how to use MongoDB Atlas Vector Search for semantic similarity matching, which completely changed how we thought about search. Keyword filters feel archaic once you've seen cosine similarity return a 99% vendor match from a plain English description.

We also learned how powerful Gemini's long context window really is in practice. Holding an entire event's worth of data — vendors, guests, budget, timeline, past conversations — in a single context and reasoning across all of it simultaneously is genuinely different from anything we'd built before. It's not a chatbot. It's a planning agent.

On the Solana side, compressed NFTs via Metaplex Bubblegum were a revelation. The cost difference between standard NFTs and cNFTs made the ticketing use case actually viable at real event scale — something we didn't fully appreciate until we saw the numbers:

$$\text{Cost per ticket} \approx 0.0012 \text{ SOL} \approx \$0.00018$$

compared to standard NFT minting which can run 100x higher. That's the difference between a gimmick and a product.

How We Built It

We started with the data model. Every feature in Com-Plan-ion flows through MongoDB — vendor profiles stored as vector embeddings, event state persisted across sessions, and Change Streams powering the live operations feed. Getting that foundation right on day one meant every integration on top of it was cleaner.

From there we built outward: Gemini for the AI planner and conflict detection, Solana for NFT ticketing and escrow payments, and ElevenLabs for the voice concierge. The frontend is a Next.js application with a sidebar that explicitly labels which sponsor powers each section — we wanted the integrations to be legible, not hidden.

The voice cloning feature was the last thing we built and the one that surprised us most. Feeding a short audio recording into ElevenLabs and hearing it come back as a responsive phone agent answering live event questions — in the organizer's own voice — was the moment the project clicked from "technically interesting" to "genuinely useful."

Challenges

Context management with Gemini was harder than expected. Keeping the full event state coherent across multiple planning sessions without hitting context limits required careful prompt engineering and selective summarization of past decisions before each new call.

Solana devnet reliability caused us real pain during the demo build. Devnet occasionally lags or drops transactions, so we had to build fallback states and loading indicators that handled failures gracefully without breaking the demo flow.

Stitching real-time updates together across MongoDB Change Streams, the Gemini planner, and the frontend required more websocket work than we anticipated. Making it feel instantaneous — rather than polling — was a problem we solved at about 3am.

The biggest challenge overall was scope discipline. Every sponsor integration had ten more features we wanted to build. Deciding what to cut, what to polish, and what to demo was harder than any of the technical problems we faced.

Built With

• elevenlabs
• elevenlabs-voice-cloning
• gemini-live
• google-gemini-api
• metaplex-bubblegum
• mongodb-atlas
• mongodb-atlas-vector-search
• mongodb-change-streams
• next.js
• node.js
• react
• solana
• solana-pay
• tailwind-css
• typescript
• vercel