# Anthony Buckley-Thorp > Anthony Buckley-Thorp is a product leader in construction technology, currently Head of Product at Prolo in London. He trained as a structural engineer — a First in Civil Engineering from Imperial College London, then skyscrapers at Arup in London and super-high-rise towers in Shanghai — and has worked in software since 2015: Flux, a Google X spin-out; co-founder and Head of Product at Helix RE (acquired by Density.io); Outer Labs, building for Google's real estate organisation; and five years as Chief Product Officer at Building Radar in Munich. After hours he builds local-AI tools and works on robotics for construction. This file is the plain-text version of https://anthonybt.com. Everything here is drawn from the site itself. ## Pages - [Home](https://anthonybt.com/): who I am, in four lines - [Timeline](https://anthonybt.com/timeline/): site → design office → product - [Product](https://anthonybt.com/product/): leadership & teams - [Projects](https://anthonybt.com/projects/): Heimdall · Mimir · Thor · JiaYou · Freyr · Völund · Huginn - [Robotics](https://anthonybt.com/robotics/): construction robotics - [LinkedIn](https://www.linkedin.com/in/anthonybuckleythorp/) - [GitHub](https://github.com/AnthonyBuckleyThorp) ## At a glance - **Background:** Structural engineer. On site and in the design office, London and Shanghai. - **Since 2015:** Product in software — Google X spin-out, co-founder, Head of Product, CPO. - **Worked:** London → Shanghai → San Francisco → Singapore → Santa Monica → Munich → London - **After hours:** Building AI apps for real problems — a speaking aid, a Mandarin tutor, a personal memory — and a robot arm. ## Product leadership I lead product in construction technology. Before that I was the customer — a site engineer, then a structural engineer on tall buildings. I've built product teams from scratch in San Francisco, Singapore and Munich, and I put as much into developing the people as into shipping the product. - **2026–now — Head of Product, Prolo.** AI-powered procurement for SME contractors. - **2021–26 — Chief Product Officer, Building Radar.** Churn from 40% to 120% NRR · reached breakeven · helped raise $7M · ARR doubled · AI-native rebuild. - **2019–21 — Client Engagement Director, Outer Labs.** Three delivery squads building SaaS for Google’s Real Estate org. - **2018–19 — Head of Product, Helix RE.** Digital-twin product 0→1; opened Singapore. Acquired by Density.io. - **2015–18 — Application Engineer → GM Product Services, Flux · Google X.** 0 → 10,000+ users; founded a services practice selling $250k engagements. ## Timeline - **2004, London — Boots on site first.** A pre-university year in industry: site engineer for Costain on St Pancras International, the HS1 Eurostar terminus. How buildings actually get built — learned on the contracting side, before the design office. - **2007, London — From Imperial to Arup.** A First in Civil Engineering from Imperial, then Arup: structural engineering on skyscrapers, starting with the 280m Pinnacle in the City of London. - **2009, Shanghai — Shanghai, where the towers were.** Moved to Shanghai because that was where super-high-rise was booming: 400m+ towers with KPF, MVRDV and Safdie. Writing code to automate design processes to keep projects on pace. - **2015, San Francisco — Turned down INSEAD for a Google X startup.** Accepted to INSEAD, then turned it down to join Flux, a Google X spin-out in San Francisco. Drove adoption 0 → 10,000+ users. - **2016, San Francisco — First team, first number.** Founded Flux’s Advanced Services practice, selling $250k engagements to Hilti, Lendlease, BCG and PwC. First time building a team and owning its number. - **2018, San Mateo / Singapore — 0→1, and an exit.** Co-founded Helix RE and led product: digital twins, ML on 3D geometry. Opened the Singapore office. Acquired by [Density.io](https://www.density.io/). - **2019, Santa Monica — Building for Google’s real estate org.** Client Engagement Director at Outer Labs. Led three delivery squads building SaaS for Google’s Real Estate org — digital twins, cost management, supply chain, workspace planning. - **2021, Munich — Five years as CPO.** Chief Product Officer at Building Radar, Munich. Turned 40% churn into 120% net revenue retention and reached breakeven, was instrumental in a $7M raise, then doubled ARR year on year and rebuilt the product AI-native. - **2026, London — Back to London.** Head of Product at Prolo — AI procurement for SME contractors. Hands-on, building it with the engineers. ## Projects Things I build after hours. All private repos except where a source link is given. ### Heimdall — in daily use A gatekeeper between cloud and local AI — sends the mechanical, token-heavy work to models running on my own machine. Coding agents are brilliant and expensive, and most of what they do is typing rather than thinking. Heimdall is the switchboard: an MCP server that discovers what Ollama and LM Studio are serving, routes the boring jobs there, and hands back propose-only diffs so nothing touches a repo without review. In Claude Code it's a /offload skill — the mechanical work goes local, and my Claude subscription goes a lot further. It has quietly become the local AI server for everything else. Freyr, my finance app, runs its AI here so the numbers never leave the laptop, and open-source coding harnesses like DeepSeek Harness and Opencode run on it fully offline. Every run gets rated — used as-is, fixed, rewritten, discarded — so the routing learns which local models actually earn their keep on this hardware. Stack: MCP, Claude Code, Ollama, LM Studio, DS4, DeepSeek Harness, Opencode. ### Mimir — in daily use A personal AI with a memory it maintains itself — a private wiki that grows and corrects itself every time we talk. Plenty of assistants remember things now, but none of them build a real working picture of my life — who people are, what I've committed to, where I'm meant to be — and use it to help before I ask. I wanted something always on, like OpenClaw, but with control over exactly how it handles my email, calendar and memories. So Mimir keeps a versioned markdown wiki per “Space” — work, home, a project — edits it live as we talk, and shows every change as a card in the conversation that I can undo. The useful part is joining the dots. It reads TfL's travel alerts against the routes my calendar says I'll take that day, and only mentions the delays that affect me. It flags the newsletters that mention topics I actually care about and quietly files the rest. Before I accept an invitation, it tells me what it clashes with. And every night it dreams: a pass that consolidates the day, links orphaned people and projects, fixes contradictions at the source, and queues at most three questions for the morning. Memory that isn't curated rots; this is the gardener. Stack: Gemini API, Gmail API, Google Calendar API, Firestore, Cloud Run. ### Thor — in initial trials A speaking aid for the iPad that listens to the conversation and suggests what to say next, so its user can take part again. I built Thor for a family member who can no longer rely on their own voice. Speaking aids tend to offer a board of stock phrases, and stock phrases don't respond to what the other person is actually talking about — by the time you've typed a real answer, the conversation has moved on. Thor listens, and borrows the best of next-word prediction: replies to what was just said, whole-sentence completions after a word or two, all in the way they'd say it. The aim is simple — to be in the conversation again, not a beat behind it. The hard part is latency, not AI. Phrases are rendered ahead of time and play in under 100 ms, suggested sentences are pre-rendered before they're tapped, common word libraries download for offline use, and if the network drops the iPad's own voice takes over so a tap never falls silent. Suggestions come from Gemini 3.8, shaped by the time of day and what Thor knows about its user, and a single keyword — doctor's surgery, chess, supermarket — grows a whole scenario library of phrases. That context syncs across phone, tablet and laptop. One firm rule: the AI only listens. It suggests, they choose, and nothing is ever said that they didn't pick. Stack: Gemini 3.8, Gemini TTS, Gemini Live API, React PWA, Cloud Run, Vertex AI. ### JiaYou — in daily use A Mandarin tutor for the Tube — spaced-repetition flashcards, short AI-generated podcasts, and a live voice tutor. I'm learning Mandarin and nothing fitted a 40-minute commute underground. JiaYou schedules flashcards with FSRS and turns today's BBC News feed — or any topic I choose — into three-to-five-minute podcast episodes that reuse my key vocab and end with a quiz. When there's signal, it holds a real-time spoken conversation that hears tone, hesitation and stress rather than transcribing them away, in scenarios built from the words I've just mastered. A kid mode swaps pub chat for playground talk, because my eight-year-old son uses it too. The live tutor uses barge-in audio, so you can interrupt it mid-sentence like a real person. Behind it sits a lesson-planner agent that tracks progress against the lesson's goals and drip-feeds context to the teacher model, so it can't accidentally give the answers away. Everything except the live tutor works offline and syncs when it can — a good reminder that the hard part of a language app is never the AI. It's the service worker. Stack: Gemini Live API, FSRS, Offline PWA. ### Freyr — in daily use Household finances as a decision engine — encrypted, local-first, and built to answer “what if?” rather than “what was?”. Every finance app wants your bank credentials and gives you a budget. I wanted a faithful record of real balances and a way to model the big calls — starting a company, moving house — as saved, comparable scenarios. It starts with a monthly snapshot ritual: net worth over time, across accounts and currencies. The interesting part is the statements. I drop in bank and card statements; parsers read them, rules file the familiar payees, and a local model served through Heimdall categorises whatever the rules can't — fully offline, so no bank data ever goes to the cloud. From that it works out the real monthly and annual recurring costs, tracks spending against budget, and projects where savings are heading. Money is stored in integer minor units with no floats anywhere, the database is encrypted with SQLCipher and unlocks with Touch ID, and nothing leaves the laptop. Also my excuse to build something fast and local in Rust. Stack: Rust, Tauri, SQLCipher, Heimdall. ### Völund — in development A desktop robot arm learning to build tessellations — show it a photo of a pattern and it assembles it from wooden tiles. I've wanted robots on construction sites since my Arup days, and the only honest way to understand why that's hard is to make a small one do a small job. Völund is a six-axis myCobot on the bench, driven from my Mac, working through a chapter-by-chapter path: a safe workspace, teach-by-hand positions, a suction cup, pick-and-place, then an overhead camera. The rule is AI only where it earns its place. Tile detection, calibration and motion are classical OpenCV and geometry; a local vision model reads the reference photo into a pattern spec and a lighter one checks each placement as it goes. Assembly is a planning problem long before it's an AI problem. Stack: myCobot 280, pymycobot, OpenCV, Gemma 4, LM Studio. ### Huginn — in daily use Menu bar screenshot markup for the Mac — Ctrl+Space, drag, draw a red box, Space to copy. Under five seconds, no toolbar. Most of my screenshots are “look at this bit” shares into Slack, Notion and email, and the built-in macOS markup takes too many clicks to get there. The Windows Snipping Tool was the benchmark: one shortcut, one drag, mark it up, gone. Huginn copies the marked-up image to the clipboard, saves a PNG and closes, all from the keyboard. The discipline was in the non-goals: no colours, no thickness, no arrows or text — a red box, a pen and undo. A native Swift app written with Claude Code in an evening, using macOS's own crosshair for capture rather than reinventing it. The fiddliest part wasn't the drawing; it was keeping Screen Recording permission across rebuilds. Stack: Swift, AppKit, SwiftUI, XcodeGen, Claude Code. Source: https://github.com/AnthonyBuckleyThorp/huginn ### AntWeb — retired My first website, hand-typed in Notepad at twelve from a library book on HTML — the first kid at school to have one. Someone had put a book about HTML in the school library, so I read it and typed the tags into Notepad until a page appeared. It went up on GeoCities, before Yahoo bought it, with a message board, a guestbook, a voting poll and a home button that was an arrow I drew myself. Chris from my class had a rival site; the poll was there to settle it. The instinct hasn't changed in thirty years: find a real problem, learn just enough to build something, put it in front of people and ask what they think. The tools have improved. The animated GIFs, I'd argue, have not. Stack: Notepad, GeoCities, Amazing Forums. ## Robotics in construction Construction is the largest industry on earth that still moves atoms by hand. A decade of construction software — including products I've led — made the paperwork faster while the site barely changed. The next decade belongs to physical AI: robots that learn from demonstration and work alongside the trades in messy, half-finished buildings. That's where I'm putting my future focus. - **The problem:** Software is only about 2% of construction spend. [As Patric Hellermann splits it](https://bricks-bytes.com/newsletter/this-unusual-partnership-signals-the-future-of-construction/): 35–40% people, as much again on materials, 10% plant and equipment. A software-only play has a tiny ceiling and leaves the actual work untouched — the prize is the other 98%, and the biggest slice is labour. - **The bet:** Bits to atoms. The venture-scale outcome is owning the site outcome: machines that do the work, not software that describes it. Robot learning has just made that possible outside the factory. - **My angle:** I've stood on the slab as an engineer and spent ten years shipping software to contractors, so I've seen how new tools really get adopted on site — and why most don't. Now I'm learning the robotics stack hands-on and working out how adoption will unfold: which tasks go first, who buys, and what it takes to enable it. ### Imperial College Robotics Summer School (July 2026 · London) Five days going deep on the stack: LiDAR and tactile sensing for unstructured environments, imitation learning for arm control, and the inverse kinematics theory underneath it all. My group worked with [Dr Edward Johns](https://profiles.imperial.ac.uk/e.johns) of the [Robot Learning Lab](https://www.robot-learning.uk/) on [Instant Policy](https://www.robot-learning.uk/instant-policy) — in-context imitation learning that lets a robot pick up a new task from a demonstration or two, with no further training ([the paper](https://arxiv.org/pdf/2411.12633) won Best Paper at the ICLR 2025 Robot Learning Workshop) — and took second prize. An excellent programme, run brilliantly by Imperial. I met interesting people from all over the world and a wide range of Imperial academics working across every area of robotics, and we visited many of their labs. I'd highly recommend it. ### Desktop masonry with a myCobot 280 (Summer 2026 · Ongoing) A six-axis arm collected in China last summer, now on the desk beside me and driven from my Mac. The proof-of-concept is tessellation: show it a photo of a pattern and it lays it out in wooden tiles. Detection, calibration and motion are classical vision and geometry; a local vision model reads the reference photo and checks each placement. Learned policies — demonstrations, simulation, sim-to-real — come next, where they beat plain geometry. ### Where it goes (The Next Decade) Cobots and mobile manipulators working beside the trades without a factory-floor retrofit: tiling, sorting, material layout, repetitive assembly. Vision-language-action models are what make a robot useful on a site that's different every day. I follow the platforms closely — Unitree quadrupeds, ARCA, NextGen Robotics — and the software and virtualised environments that run them.