
Mateusz
Bochyński
Customer discovery · Feature validation · Rapid prototyping
I work where the product meets the customer. At Binderless I reached 139 construction companies in 12 weeks, took 35 through discovery and turned what I heard into a prototype the CEO and CTO saw four weeks later. Before that: my own paid product, and an AI assistant that took the guesswork out of two hotel receptions.
- Reached 139 construction companies in 12 weeks: 477 cold calls, up to 68 a week, 213 live conversations, 35 companies taken through discovery, demos and onboarding for paying accounts.
- 1 in 6 cold-called companies booked a discovery call (22 of 125). Written follow-up typically the same day, median 3.4 hours after the meeting.
- Validate new platform features with customers before release (AI agent, AI reports module): calls and demos with active and paying users; findings go back to product.
- From customer calls to a working prototype in four weeks, after hours: IFC-Agent answers questions about a building model, shows the source of every number and highlights elements in 3D. Built the case from 59 recorded customer calls (66 signals, 32 companies), demoed to the CEO, CTO and team.
- Built our post-meeting workflow: transcript to MEDDPICC note in Pipedrive to branded summary deck and follow-up email, now our standard after every customer meeting. Behind it: a 3,900-note Obsidian knowledge graph my AI agents read and write.
- Solo from discovery to paid launch: product, UX, architecture, legal pack and Stripe payments. v1.0 live October 2026, first paying customers within days of launch.
- Own the business side too: pricing and unit economics per generation, cost and margin per AI provider, refund policy, EU terms and privacy, analytics, founder-led content.
- Positioning built with designers and studios: controlled iteration instead of one-shot rendering. One model, many variants: materials, seasons, light, camera preserved.
- Built for a reception with constant turnover and 304 procedures, where new hires guessed rather than ask the manager a “stupid question”. Atlas takes the fear out of asking: a step-by-step answer with the source procedure and a warning when a decision could cost the hotel.
- Result: receptionists stop improvising, wrong decisions stop at the question instead of after the guest complains, and the manager is no longer the only source of answers.
- Designbotic · QA Tester 2025 – 2026 · Revit tools and release QA, bug reports with repro steps.
- ROARK Studio · Architect 2024 – 2026 · BIM documentation, AutoLISP drafting automations.
- TMRW · BIM Modeler 2022 – 2023 · Models for Foster + Partners, Zaha Hadid and SOM projects.
Three things I built, and what they changed.

IFC-Agent
Prototype module for Binderless. Ask a building model a question, get the number, its source and the elements lit up in 3D.
- Problem
- Contractors check material requests and RFIs against the BIM model by hand. Viewers show geometry, not answers, so the model sits unused after design.
- Built
- Turns an IFC file into a queryable database, answers in plain language, shows the query behind every figure and highlights the matching elements in 3D. 198 automated tests; a 230 MB model analysed in 15 seconds.
- Result
- From customer calls to a working demo in four weeks, after hours. Shown to the CEO, CTO and team, now validated with customers. Evidence: 59 calls, 66 signals, 32 companies.

Proposition Studio
AI image iteration for architects. Change one part, keep the design.
- Problem
- AI renders are easy. Changing one thing without wrecking the design, keeping the camera, the massing and the direction the client approved, is not.
- Built
- Hosted app: references, labels and masks, generation, variant compare, history. Cloudflare Workers, Postgres, R2, fal.ai. Legal pack, Stripe credit packs, USD pricing.
- Result
- v1.0 live October 2026, first paying customers within days of launch. Solo: discovery, product, UX, architecture, legal, payments.

Atlas Hotel Assistant
Front-desk AI for Hotel Faros and Hi Hotel in Gdańsk. In production since June 2026.
- Problem
- A reception with constant turnover and 304 procedures. New hires guessed rather than ask the manager a “stupid question”; errors surfaced after guests complained.
- Built
- An assistant over the procedure base: step-by-step answers with the source procedure and a warning when a decision could cost the hotel. Admin dashboard, analytics, feedback loop. Next.js, Convex, Gemini.
- Result
- Receptionists stop improvising and ask Atlas. Wrong decisions stop at the question. 6 in 10 questions answered instantly, no lookup needed.
A three-month trial: one measurable outcome, picked together and delivered; then you decide whether we continue.
Customer calls, an evidence ledger with quotes, a go / no-go memo your team can act on.
I sit in on discovery, map the workflow and come back with a working prototype your customers can click.
Transcript to qualification note to follow-up, built into your stack and kept by your team.
Three months, one measurable result, then you decide. · m8@mateuszbochynski.com · +48 732 190 799