AI Ask is a grounded engineering assistant built on top of your tenant data — soil logs, lab results, calibration records — and the rule packs you work to. Every claim points to a numbered source you can open. It also computes with the same code-checked engines as the rest of the platform, and acts on your say-so: set up a project, import a pasted borelog, build a custom bore-log template, or draft a client proposal.
Bolt-on chatbots can only talk. AI Core is wired into your tenant — it reads across every borehole, SPT record, lab result and the immutable audit log, runs the same code-checked engines the rest of the platform uses, and acts on your instruction: set up a project, parse a pasted borelog, build a bore-log template, or draft a client proposal. You stay in the chair — it never signs anything.
And it shows its work. Every answer decomposes into claims, and each claim is anchored to its source — a borehole number, a lab test ID, a governing clause. Strip the citations away and the answer dissolves. That is the point: defensible by construction.
These aren’t canned demos — every example below maps to a real tool AI Core calls against your own data, live.
No copy-paste, no re-keying. AI Core reads across your boreholes, stratigraphy, SPT, samples, lab results and the audit log — and answers from the actual records, cited.
Hand it the numbers and it computes with the same golden-tested engines as the platform — bearing capacity, slope, seepage, piles, liquefaction and more — and every result cites its governing clause. It never calculates by hand.
Tell it what you want and it builds it — provision a project, parse a pasted borelog into real boreholes, generate a bore-log template to your firm’s format, or draft a client proposal grounded in the job. You review and approve; it never signs.
AI Core sees the live state of every project — what’s pending, what changed, what’s missing — and answers from the tamper-evident audit chain. Every question it answers is itself logged.
Pick a question to see a sample answer. Notice every claim ends in a citation — a standard clause, a borehole, an analysis. Strip the citations away and the answer dissolves. That is the point.
What is the net safe bearing capacity for a 2 m wide strip footing at 1.5 m depth in BH-102?
Answers decompose into claims. Each claim links to its source — a borehole number, a lab test ID, a standard clause. Click any claim to open the underlying record.
Each ask is written to the audit chain alongside the answer it produced, the model that generated it, and the citations it relied on. You can reconstruct the conversation a decade later.
IS 1893, ASTM, Eurocode 7, BS 5930 — the rule packs ship as structured context so the model classifies and compares using the same rules an engineer would.
The primary engine for analytical answers, classification, engineering narratives, and long-document Q&A — pulling claims out of multi-hundred-page geotech reports without truncating the source. Picked for accuracy and grounding behaviour on long, structured prompts.
Reads field photos and proposes USCS classification. Used inside AI Ask when you attach an image, and inside the mobile field app for the capture flow and paper-log OCR.
A low-latency model for quick checks — validator explanations and input sanity checks — where a fast response matters more than depth. Same Bedrock region, same no-training guarantee.
All inference is invoked through Bedrock in the customer’s AWS region. No data is logged outside the region. No foundation model is trained on your data.
Every user-facing AI path runs on open-weight foundation models hosted in your own AWS region through Bedrock — no customer data leaves the region, and no foundation model is ever trained on your data.
1-month free trial. Bring an AGS file and a couple of lab results — you will know in five minutes whether this is for you.