LLMs vs BV360 Pre-Flight
Can a generic AI chatbot review architectural drawings against adopted city code?
Short answer
A generic LLM is useful for drafting emails and summarizing a published code page you paste in. It does not reliably read CAD dimension strings off a plan-set PDF, and it does not hold the adopted amendments for each suburb. Pre-Flight uses the same vision and language models, but binds them to a specific AHJ code library and a verify step so the output is cited, not confident-sounding.
Why you can't just upload your PDF into an AI chatbot
Pre-Flight is built on AI language models too — that's the point. An off-the-shelf AI chatbot has no memory of your city's code. Pre-Flight puts three guardrails around the model:
- 1
An adopted-code library per city
Pre-Flight carries the actual municipal code for the city you pick — Skokie Chapter 118, Darien Title 5A, and more. The reviewer is instructed to cite only sections literally present in that jurisdiction's code reference and to return nothing rather than invent a section. Upload the same PDF into a generic AI chatbot and it will recall a generic IRC setback that may not apply.
- 2
A verify step before the review runs
Pre-Flight reads the cover sheet and site plan with vision, pre-fills the project data, and asks you to confirm the extracted dimensions before the review. A generic AI chatbot gives you one confident pass with no way to check the numbers it read.
- 3
Structured findings, not prose
Output is a pass / flag / critical finding per dimension with the exact municipal section cited, and local amendments override the base model code (e.g. Skokie's Type-K copper requirement). A chatbot hands you a paragraph; Pre-Flight hands you redline notes.
At a glance
| Capability | Generic LLM | BV360 Pre-Flight |
|---|---|---|
| Reads your PDF dimensions | Unreliable — plausible but unverified | Vision pass on cover + site plan, then you verify |
| Which code? | Whatever it remembers or browses | Adopted library for the city you pick |
| Hallucinated sections | Common — generic IRC that doesn't apply | Instructed to cite only sections in the city's code reference; returns nothing rather than invent |
| Local amendments | Missing | Applied (e.g. Skokie Type-K copper over PVC) |
| Accessory / garage rules | Usually uses the principal-building yards | Judge detached garages against the accessory rules (3 ft lot line / 5 ft alley) |
| Output | Prose | Findings you can take to redlines |
| Permit? | No | No. The city still decides. |
Worked example
Illustrative: a Skokie R-1 addition proposes a 2.75 ft side yard. A generic chatbot may recite a generic model-code setback that does not apply. Pre-Flight checks the proposed yard against the Village of Skokie R-1 standard (two side yards, neither less than 6 ft) and returns a flag with the section citation, so you know before you file.
When to use which
Use a generic AI chatbot to write the resubmittal letter. Use Pre-Flight to see whether the sheet fails Skokie / Darien / Overland Park / KCMO before you file.
FAQ
Does Pre-Flight replace the plan examiner?
No. Pre-Flight is a builder pre-submittal check. The authority having jurisdiction still makes the final decision.
Pre-Flight uses an LLM too. Why can't I just use any AI chatbot?
Pre-Flight runs the same vision + language models, but it wraps them in three things a raw chatbot does not have: (1) an adopted municipal code library per city that the reviewer is forced to cite from — it is instructed to never invent sections; (2) a verify step where you confirm the extracted dimensions before the review runs; and (3) structured pass/flag/critical findings with exact municipal citations you can take to redlines.
Do I need a credit card to try it?
No. Two lifetime reviews are free with no credit card.