Production AI system for aircraft sales, automating inventory accuracy, lead generation, and quote delivery for aviation brokerages.
Aircraft brokerages operate on thin margins with high-touch sales processes. Inventory changes daily, quotes require manual research, and leads slip through the cracks when staff are busy with existing customers.
The customer, an aircraft sales brokerage, was managing inventory accuracy manually, responding to leads when time allowed, and delivering quotes through time-intensive human processes. They needed a system that could operate 24/7 without sacrificing accuracy.
As Founder & CEO, I owned everything: product strategy, technical architecture, AI implementation, frontend development, and customer acquisition. With a small team of 2-4, I was hands-on across every layer of the stack.
This meant making tradeoff decisions daily between building new features and maintaining what was already shipped, between perfect solutions and good-enough-to-learn solutions.
Every startup is a series of bets. Here are the ones that shaped EA Aviation:
Chose Voiceflow over building custom, trading flexibility for speed-to-market. This let us ship and iterate with real users faster.
Selected Anthropic's Claude for response quality and safety. In aviation, accuracy matters more than creativity.
Met the customer where they were. Their inventory lived in spreadsheets, so we integrated directly rather than forcing migration.
Resisted the temptation to build a horizontal platform. Went deep in one vertical to prove the model before expanding.
The system prioritizes accuracy over creativity. Inventory lookups use code-based API calls to Google Sheets rather than letting the LLM guess. Claude handles natural language understanding and response generation, but never fabricates data.
Early stage, but real results with a live customer:
The EA Aviation architecture proved reusable. Once the core pattern worked, we applied it to adjacent verticals with different data sources and user needs.
AI-powered dealership platform handling inventory lookup, lead generation, and automated scheduling.
Municipal services directory for Cincinnati. Gives residents fastest links and information for city services.
Strip features to ship. The original vision was far more complex. We cut 60% of planned features to get something real in front of users. Every cut hurt, but every cut got us closer to learning.
Trust and accuracy beat features. Users don't care about clever AI tricks. They care about correct inventory and reliable quotes. We optimized for boring reliability over impressive demos.
Real user feedback is the only feedback that matters. Every assumption we had was wrong in some way. Production users taught us what actually mattered.
Technical constraints shape the product. What felt like limitations (Google Sheets, Voiceflow's boundaries) actually forced better decisions. Constraints are features.
Start even smaller. We could have validated the core value proposition with half the features. The fastest path to learning is the smallest possible thing that creates real value.
I'm happy to walk through the technical architecture, decision tradeoffs, or lessons learned.
Get in Touch