Nicer

I am the first engineer at Nicer. I wrote the first line of code and built the technical base the product still runs on.

Founded

2023, United States

Funding

$2M pre-seed, led by Trip Ventures, announced February 2024

Board and advisors

Include the Global Director of Travel at Microsoft, the President of Global Travel Collection, and an Operating Partner at SoftBank

Partners

IATA · Internova · Microsoft · OpenAI · Curated · Mr & Mrs Smith

Everything above is published by Nicer. The operating numbers are the company’s, not mine to put on my own site.

The problem it works on

Elite travel advisors are very good and very scarce. Each one can only handle so many clients, because planning a real trip is hours of research, supplier calls and itinerary assembly per booking.

The thesis is to raise that ceiling with AI and automation without removing the human — keeping the judgement and the supplier access that make an advisor worth paying for, and deleting the parts that are just typing.

What I built

The platform under the advisor portal — client chats, client profiles, payments, hotel and flight booking, and an itinerary builder whose output ships as an app, an email, a PDF or a web link.

The integration layer. A long list of travel and technology providers, each with its own API shape, data model, authentication scheme and failure mode, normalised into something the product can treat uniformly. This is the least glamorous and most load-bearing part of the system: travel inventory is famously inconsistent, and every provider is inconsistent differently. I wrote about what that actually takes.

The AI layer, on AWS, that turns unstructured requests into structured trip work.

Everything underneath it — infrastructure, observability, deployments and production operations. On a founding team, ownership is not a claim about scope. It is a description of who gets paged.

What it taught me

Being the first engineer means every early decision is either load-bearing or expensive to reverse, and you make them with no traffic to learn from.

The ones that held up were the ones where I optimised for how quickly the next thing could be changed, rather than for how correct the current thing looked.