Back-end and mobile engineer to make a live document engine compound and learn (5 weeks, start now)

We run companies in adventure travel and operate on all seven continents. Antarctica and the Arctic, ski and heli-ski, boats, remote lodges. The documents those trips run on are itineraries, charter brochures, regional overviews, vessel specifications, packing lists and quotes. Fourteen types in all. We have an engine that generates them. A user picks a region on a map and a document type, gives the engine six pieces of information about the trip, who, what, when, where, why and how, and it drafts the document in our house voice, pulls photographs from our library, runs automated quality checks and exports a PDF. Every document has an edit tab: change the text, swap a photograph, ask for a rewrite in a chat bar. It is hosted and the team uses it on real client work. The product is nearly there. The front end works. We need a back-end and mobile engineer, ideally one person doing both, and we will also consider back-end only. PHASE ONE. THE FIRST TEN DAYS Get the back end fully functional. Four things. 1. Placing the six inputs correctly. This is the biggest one. The user tells us who, what, when, where, why and how. During ingestion the generator does not reliably put each of those where our framework says it belongs, so a draft can read well and still be structurally wrong. 2. Map routing and itinerary generation. Routes and day-by-day itineraries come out wrong often enough that someone checks every one by hand. 3. Text and image placement, copy and design consistency. A photograph lands beside a paragraph it does not match. Layout and copy drift from our house design. 4. Ingestion and export. Source PDFs and spreadsheets lose content on the way in. Exported PDFs do not always match what is on screen. PHASE TWO. THE REMAINING WEEKS Work directly with our team as they run real documents through it. Take their feedback, fix what breaks, harden it, pressure test it, and build what the use exposes. Underneath all of it is the thing we care most about: every document a human approves should make the next one better, and right now that loop is weaker than it should be. SCOPE This is a working MVP. Real client documents come out of it every week. We are not looking for a rewrite, a reorganisation, or a move to a different framework. We are looking for someone to bring their own experience and take ownership of the model underneath: organise it so it runs more efficiently and gets better at compounding and learning over time. THE STACK Python for ingestion. TypeScript and Node on the server. An LLM API for generation. Leaflet for the maps. Next.js and React on the front end, already built. Hosted on Vercel. WHO YOU WORK WITH You work directly with the owner and founder, Stein Retzlaff, and the team. https://steinretzlaff.com https://instagram.com/steinretzlaff Our documents go to private clients, so discretion is part of the job. BUDGET AND TIMING $1,500 across five weeks, in two milestones. $500 when the back end is working. $1,000 for the rest: upkeep, hardening, pressure testing and whatever the users surface. We want to start immediately. There is a lot more to share once we talk, and more work after this if it goes well. TO APPLY Start your proposal with the word ANTARCTICA. Then answer three questions. 1. How do you make this compound and learn? 2. How do you learn from the most recently approved and referenced document? 3. We hand you a PDF full of mixed material: text, photographs, a map, a schedule. How would you turn it into a new, well-designed document instantly? Describe the approach, not the tools you would reach for. We look forward to working with you.