Bild AI: Product, team and how it was built
Bild AI (AI takeoffs for Division 8 (doors, frames and hardware): reads blueprints and specifications, builds door and hardware schedules, and exports the bid to the distributor's quoting system, United States): Sold to Division 8 distributors and estimators, November 2026. Evidence grade E: no revenue figure was checked. Team: 2 co-founders at the February 2025 launch; 9 people on the Y Combinator company page in November 2026, after a seed round led by Khosla Ventures. Launched February 2025; still running.Added 8 November 2026
- Outcome
- Sold to Division 8 distributors and estimators, November 2026
- Evidence
- E no revenue figure was checked
- Launch
- February 2025
- Status
- Still running
- Team
- 2 co-founders at the February 2025 launch; 9 people on the Y Combinator company page in November 2026, after a seed round led by Khosla Ventures
- Outside funding
- Seed round led by Khosla Ventures, announced July 2025 (US$3.1M as stated by Signalbase and Dev Curation; the company's own post gives no amount)
- Main channel
- Not stated
What is not verified
No revenue figure, price or customer count was found. The Y Combinator pages, the Launch HN post, the company blog and the funding notes give none, and no Bild AI profile is in the TrustMRR slug list. That the product is sold rests on the company's own words ("clients", "customers", "Our client network"), its logo strip and a PropTechBuzz news note of 3 July 2025 ("The startup operates on a subscription-based revenue model"); no named customer has confirmed it in a source that was read. The US$3.1M seed amount comes from Signalbase and Dev Curation; the company's own announcement gives no amount. The start year is 2024 on the Y Combinator page and February 2025 in Dev Curation. A reader could look for a customer case study that names the distributor, a price in a contract or a public tender, or a founder interview with figures. This story is grade E: the business is real and its product, price and customers are described from public pages, but no revenue figure was checked. Aiolum shows it as an idea worth studying, not as proof that it makes money.
Bild AI is a San Francisco company that sells takeoffs, the count and measurement of every material item in a building's drawings, which a supplier needs before it can price a bid. It now works only on Division 8, the section of the standard construction specification format that covers openings: doors, frames and door hardware. It reads the door schedule (the table in the drawings that lists every door with its size, material and fire rating) and the hardware sets (the lists of hinges, locks, closers and other parts that belong to each type of door), and exports them to the distributor's quoting system.
Roop Pal, who built computer vision models at Google and Waymo, and Puneet Sukhija, who had built houses in Canada, started it in 2024 and launched it in February 2025 in Y Combinator's Winter 2025 batch. In their Launch HN post they wrote that they were "live with customers". In July 2025 the company announced a seed round led by Khosla Ventures and a model trained only for Division 8; Signalbase reported the round as US$3.1M. In November 2026 the Y Combinator page lists 9 people, and the home page shows the logos of 18 door and hardware distributors under "Trusted by".
No revenue figure, price or customer count was found, so this story is grade E: the headline is a status, that the company sells to Division 8 distributors, as the company itself states. The bid increases that the company reports for its clients are customer outcomes, not revenue.
How much revenue does Bild AI make?
Each row shows the evidence grade and the source numbers from the list at the end of this page. The grades are explained in How Aiolum verifies.
| Fact | Value | Grade | How we know | Sources |
|---|---|---|---|---|
| Status, November 2026 | sold to Division 8 distributors and estimators, as the company states | C | Home page: "Trusted by Division 8 distributors & estimators nationwide" and "Our client network across North America". Y Combinator page: "clients have bid 50% more with smaller estimating teams." PropTechBuzz (3 July 2025): "Bild’s early customers include suppliers in the framing, flooring, and door industries" and "The startup operates on a subscription-based revenue model." No source states a price, a contract or a named paying customer. | [1, 2, 11] |
| Customers named by the company | 18 distributor logos, among them ARJ Architectural, Schiller Hardware, Willis Klein Commercial and NY Fireproof Door | C | Logo strip on the home page under "Trusted by Division 8 distributors & estimators nationwide". No statement by any of these companies was found. | [1] |
| Founders | Roop Pal and Puneet Sukhija | C | Y Combinator page and the Launch HN post: "We're Roop and Puneet, the founders of Bild AI". The company blog says they met at Hack for Social Impact. | [2, 4, 5] |
| Company start | 2024 | C | Y Combinator page: year founded 2024. Dev Curation (15 October 2025) writes "Founded Feb 2025", which is the month of the public launch. The Internet Archive copy of bild.ai from 20 December 2024 is a page that offers the domain for sale, so no earlier product ran on this domain. | [2, 8, 9] |
| Launch | February 2025 | C | Y Combinator Launch post dated 11 February 2025; the company blog (2 July 2025): "Since launching in February, our models have analyzed blueprints across 6 countries". | [3, 5] |
| Team | 2 founders at launch; 9 people in November 2026 | C | The Launch HN post is written by the two founders. The Y Combinator page gives a team size of 9 (read 8 November 2026). Signalbase listed "2-10 employees" on 3 July 2025. | [4, 2, 7] |
| Funding | seed round led by Khosla Ventures, with Mission Street Capital, Ryan Sutton-Gee and Ooshma Garg | C | Company blog, 2 July 2025, with no amount. Signalbase (3 July 2025) and Dev Curation (15 October 2025) give US$3.1M. | [5, 7, 8] |
| Prices | not published | C | bild.ai/pricing answers 404; see the pricing table. | [1] |
| First archived product page | 21 February 2025: "Understand blueprints with AI" (the page description) | B | Internet Archive copy: material cost estimation, code compliance checks for permits and blueprint accuracy for any project, with "Featured on Y Combinator". Not yet aimed at Division 8. | [9] |
| What the company says clients gained | 10x more projects (February 2025); 50% more bids (November 2026); 40-60% more bids per estimator (November 2026) | C | Y Combinator Launch post: "Our first clients are using Bild AI to bid on 10x more projects". Home page: "40-60% more bids per estimator", with the source given only as an unnamed Division 8 distributor. These are customer outcomes, not Bild AI's revenue. | [3, 2, 1] |
| Quoting systems it exports to | Comsense, PortalPro, AVAware, TOAD, eMullion, SFH; the Y Combinator page also names ProTech | C | Home page section "Push straight into your quoting system" and the Y Combinator description. | [1, 2] |
| AI models | its own computer vision models, chained with vision-language models; provider not named | C | Launch post: "applies computer vision models and large language models". Launch HN reply by Roop Pal: "You can chain those CV models with VLMs". No page names a model provider. | [3, 4] |
| Time from launch to the November 2026 status | 21 months | calc | Calculated: February 2025 to November 2026. | [3, 1] |
How to read this table: a grade C value is the company's own statement (its site, its Y Combinator pages, the founders' Launch HN post) or a news note about its announcement. A grade B value is an independent record: dated Internet Archive copies of its home page. An "X" value was computed on this page. No value is grade A.
| Plan | Per month, billed yearly (US$) | Per month, billed monthly (US$) | Limit |
|---|---|---|---|
| Division 8 takeoffs | not published | not published | by upload form and contact; no price on the site |
On 8 November 2026 bild.ai/pricing answers 404 and the site has no price on any page that was read. Buyers start with "Upload your plans", a form that says "We'll respond frighteningly fast", or the address contact@bild.ai. PropTechBuzz wrote on 3 July 2025 that "The startup operates on a subscription-based revenue model"; no source gives a price, a plan name or a billing period.
When did Bild AI launch, and how did it grow?
From a for-sale domain in December 2024 to a product for one construction trade in November 2026. Each entry carries its grade.
- 20 December 2024Bbild.ai shows a page that offers the domain for sale. [9]
- 11 February 2025CThe Y Combinator Launch post says Bild AI reads blueprints for "flooring, doors, hardware and more" and that its first clients are flooring suppliers. [3]
- 21 February 2025BThe home page description reads "Understand blueprints with AI", and the page offers estimation, permit code checks and accuracy checks, with an upload form. [9]
- 27 February 2025CThe founders' Launch HN post gets 98 points: "We're currently live with customers but are only able to serve a subset of trades accurately right now." [4]
- 2 to 3 July 2025CThe company announces a seed round led by Khosla Ventures and introduces "Bild AI D8, a model built and trained specifically to serve the Division 8 industry"; Signalbase reports US$3.1M. [5, 6, 7]
- 8 November 2026CThe site speaks only to Division 8 distributors and estimators, shows 18 distributor logos, and the Y Combinator page lists 9 people. [1, 2]
Why did Bild AI work?
Four causes, in the order they mattered. Each one is a reusable lesson.
One trade instead of all trades
In February 2025 the product read blueprints for flooring, framing, doors and permit checks. By July 2025 the founders had trained a model only for Division 8, and by November 2026 the home page sells nothing else. One trade has the same document types and symbols on every project, so a narrow model can be checked against a known answer.
Many small models instead of one large model
In the Launch HN post Roop Pal wrote that they use "a suite of specialized machine learning models for specific blueprint comprehension tasks", such as detecting floor areas or counting doors, and compose them, as perception teams do for self-driving cars.
Output in the format the customer already imports
The home page says the export is "Sized, split, and formatted to import" into the quoting systems that door distributors run, such as Comsense and AVAware, so the estimator does not type the lines again.
Checks that an estimator would do by hand
The software compares each opening across the plan, the schedule and the specification, and lists added, removed and changed doors between drawing versions, which helps the estimator write a <dfn>request for information</dfn> (RFI), a formal question to the architect about a conflict in the drawings.
How it was built
These details come from the company's site, its Y Combinator pages, the founders' Launch HN post and the page source of its website. They are grade C unless marked. [4, 1, 2, 9, 6]
- Computer vision models for detection and measurement tasks on drawings, chained with vision-language models, for example to get the drawing scale right (Launch HN).
- A web application at app.bild.ai where customers upload plans, edit openings and export the takeoff, with an assistant called Bildy for bulk edits in plain English.
- Exports in the import formats of door-distributor quoting systems: Comsense, PortalPro, AVAware, TOAD, eMullion and SFH.
- The 2025 site and the blog were built with Framer; the current home page is served by Ploy and built with Astro; both load PostHog and Google's gtag script, and the 2025 site loaded a LiveChat widget (page source, grade B).
- No page names the language model provider or the cloud provider.
Tools named in the sources
Each one opens its page in the Tools tab, with every story that uses it.
Who tried this before Bild AI, and why did they fail?
Three companies in our database of failed startups promised to make building cheaper and faster with technology and failed. Two of them, Katerra and Veev, also had Khosla Ventures among their investors. The values below are from that database and are not independently verified.
Katerra
The promiseOne company for the whole building process, from design and factory-made parts to construction.
Why it diedIt built factories and supply chains before it had enough customers, then bought other companies and went bankrupt.
Veev
The promiseFactory-made wall panels with built-in wiring and plumbing, so homes could be assembled faster and cheaper.
Why it diedIts factory needed large and steady demand, and higher interest rates cut home demand while it still had high fixed costs.
Social Construct
The promiseFaster and cheaper housing through a technology-driven way of building.
Why it diedIt expected to scale faster than construction rules and the pandemic allowed, and ran out of money.
The failed startups come from our database of 1,749 dead startups, retold here in our own words. Its entries are AI-assisted summaries of public sources, so these cards carry no evidence grade.
What changed
The three predecessors tried to change the building itself: factories, panels or a new construction method. Katerra raised US$2.6B and Veev US$647M, as our database of failed startups records, and all three needed buildings to sell before the investment paid back.
Bild AI changes nothing on the building site. It reads the drawings that a distributor already receives for every bid, a task that computer vision and vision-language models can now do for one narrow trade. It sells to an existing step of an existing business, so a customer can see the result on the next bid. The open question, which no source answers, is how much distributors pay and whether they keep paying.
Can one person copy Bild AI now?
Not as a solo project, and not without trade knowledge and training data.
The founders combined computer vision experience from Google and Waymo with years of house building, joined Y Combinator and raised a seed round. A Division 8 takeoff needs models trained on real drawings, exports for several quoting systems, and estimators who trust the result. The method, one trade and one document type, can be copied in other trades by someone who knows that trade's drawings.
Where it is still open: Other construction trades with their own schedules (windows, plumbing fixtures, electrical devices, fire protection), and suppliers in countries with other drawing standards.
- 1Pick one trade whose estimators count items on drawings by hand for every bid.
- 2Train small models for each detection task (doors, tags, schedules) and check them against finished takeoffs from real customers.
- 3Export in the exact import format of the quoting systems that trade already uses.
- 4Add the checks an estimator does by hand: conflicts between plan, schedule and specification, and changes between drawing versions.
What could go wrong for Bild AI?
The risks are the mirror image of the causes above.
What is not verified
No revenue figure, price or customer count was found. The Y Combinator pages, the Launch HN post, the company blog and the funding notes give none, and no Bild AI profile is in the TrustMRR slug list. That the product is sold rests on the company's own words ("clients", "customers", "Our client network"), its logo strip and a PropTechBuzz news note of 3 July 2025 ("The startup operates on a subscription-based revenue model"); no named customer has confirmed it in a source that was read. The US$3.1M seed amount comes from Signalbase and Dev Curation; the company's own announcement gives no amount. The start year is 2024 on the Y Combinator page and February 2025 in Dev Curation. A reader could look for a customer case study that names the distributor, a price in a contract or a public tender, or a founder interview with figures.
Trust signals on the company's pages
The 18 distributor logos under "Trusted by Division 8 distributors & estimators nationwide" carry no quote, date or link, and no statement by those companies was found. A section headed "Customer success story" shows a video titled "Bild AI — Division 8 Takeoffs Demo" and names no customer. The bid increase of 40% to 57% year on year is attributed to an unnamed "Division 8 distributor, $100M+ annual revenue" (the home page's words). The claimed gains changed over time: "10x more projects" in February 2025, "50% more" on the Y Combinator page and "40-60% more bids per estimator" on the home page in November 2026. The February 2025 home page showed "95% Model Accuracy" and "6 months Average Time Saved" without a method. The November 2026 home page also states "98+ sites in one year" under "Our client network across North America", with no list of the sites and no method.
Team size and funding
Two founders launched the product in February 2025. The Y Combinator page lists 9 people in November 2026, after the seed round led by Khosla Ventures. Under the site's rule for funded teams of 6 to 10 that were started by 1 to 5 people, the story keeps the label "Small team".
Competition
On this site Tyce reads all bid documents for general contractors and Blueprint Crusher parses bills of quantities, while Bild AI works for one trade, Division 8 distributors, and exports to their quoting systems.
Dependence on model providers
The Launch post says Bild AI uses "computer vision models and large language models", and Roop Pal wrote on Launch HN that they chain their own computer vision models with vision-language models. No page names the provider of the language models. The detection models are the company's own, so a change at an outside provider would reach only part of the product, and no source says which part.
Sources
The grade next to each source is the best grade that the source can support. "Read in full" means that the text of the page was downloaded and the numbers were checked in it.
- C Bild AI home page, read 8 November 2026Read in full, including the page source. Source for the Division 8 product, the logo strip, the ROI section, the quoting systems and the site tools. bild.ai/pricing and bild.ai/about answered 404 on the same day.
- C Y Combinator company page: Bild AI, read 8 November 2026Read the company data embedded in the page. Source for the description, the founders, the year founded (2024), the team size (9), the batch (Winter 2025) and the Khosla Ventures backing.
- C Y Combinator Launch: Bild AI – Read and Understand Blueprints using AI, 11 February 2025Read in full. Source for the first product scope, the first clients (flooring suppliers), the "10x more projects" claim and the computer vision and language models.
- C Launch HN: Bild AI (YC W25) – Understand Construction Blueprints Using AI, 27 February 2025Read the founders' post and Roop Pal's replies. Source for the founders, the specialized-model approach, the VLM chaining and "live with customers".
- C Bild AI blog: Time to Bild, 2 July 2025Read in full. Source for the seed round and its investors, the February launch, "6 countries", "more than half a dozen construction customers" and how the founders met.
- C Bild AI blog: AI for Division 8 (Doors, Windows), 2 July 2025Read in full. Source for the Bild AI D8 model and the export to Comsense, Emullion and Avaware.
- C Signalbase, 3 July 2025: Blueprint for Success: Bild AI Raises $3.1M to Revolutionize Cost Estimation in ConstructionRead the free part in full. A funding note that quotes the company's announcement; source for the US$3.1M amount and "2-10 employees".
- C Dev Curation, 15 October 2025: Company Spotlight: Bild AIRead in full. A newsletter profile; source for the US$3.1M seed in July 2025 and its "Founded Feb 2025" line.
- B Internet Archive copies of bild.ai, 20 December 2024 and 21 February 2025Read both copies in full, including the page source. The December 2024 copy is a domain-for-sale page; the February 2025 copy is the first Bild AI product page.
- C Database of failed startups, local exportLocal export. AI-written summaries, not independently verified. Source for the failed predecessors.
- C PropTechBuzz, 3 July 2025: Construction Startup Bild Secures $3.1 Million to Streamline Housing Development with AIRead in full. A news note on the seed round. Source for "Bild’s early customers include suppliers in the framing, flooring, and door industries", "The startup operates on a subscription-based revenue model" and "Currently operated by just the two co-founders".
Run this company? Send corrections or a verified figure to aiolum@livyden.com