Engineering inputs, not just text
Drawing sets and GD&T. Tolerance stack-ups. Tables inside scanned PDFs. Workbooks whose formulas reference three other workbooks. The commercial terms buried on page 40 of the purchasing agreement.
AI quoting for engineered parts and custom systems
Time kills deals. Nyotta defeats time.
Defensible quotes in minutes, not weeks.
Nyotta turns drawings, specifications, and commercial requirements into structured scope, BOMs, and cost estimates. Your engineers review the evidence, resolve open questions, and approve the quote.
RFQ-2026-0417Rev DIn review
Part
HX-214 thermal plate
Ingest ✓Read ✓Resolve ✓EstimateApprove
Documents in package48
Requirements extracted214
BOM lines structured312
Should-cost estimate per unit$248.40
Open exceptions for review7
Source links attached533
Received 08:12 · running 1 h 01 mDue Fri 12 Sep
Awaiting engineer approvalReview
SOC 2 Type II
Audited security controls
Source-linked outputs
Every line cites its sheet and zone
Approval before write-back
Nothing reaches the ERP unapproved
No data prep
STEP, IGES, DWG, DXF, PDF, XLSX, DOCX, native CAD, scans... Work on the package as it arrived.
The number is a judgment call. Getting to the point where you can make it takes weeks.
Nyotta reads the whole package first. Every sheet, specification, workbook, and clarification. Then it resolves the revisions, structures the variant-level BOM, builds the should-cost line by line, and writes the exception list. Reconciliation starts on day one, and every number arrives with the sheet and zone behind it, for you to judge.
The package is fragmented
A 2D drawing set, a STEP AP242 model, a 14-tab commercial workbook, and eleven emails of clarifications. Nothing in the package references anything else in it, and the revision that governs is rarely the one on top.
Exceptions hide in the revs
Rev C in the print. Rev D in the spec.
Knowledge does not carry forward
You quoted a near-identical bracket in 2023. The assumptions that made that number defensible live in a local workbook.
Two kinds of hard quote.
High-variant component programs, and systems that have never been built before. One is priced against a variant matrix, the other against a scope nobody has built yet. Same platform.
Automotive Tier-1Explore Automotive Tier-1
High-variant programs, resolved to the part number.
Read the drawing set against the variant matrix, carry GD&T and datum callouts into the BOM, and price every variant without rebuilding the estimate four times.
Drawing set and variant matrix intake01
Variant-resolved BOMs02
Should-cost and manufacturability (DFM)03
Engineering change (ECN) impact on a live quote04
Custom AutomationExplore Custom Automation
One-off systems, priced before anyone has built one.
Whether the cell can meet the spec, what it takes to build, what it costs, and what margin survives. Applications and sales working from the same numbers instead of two spreadsheets.
Scope and technical feasibility01
Equipment BOM and vendor quotes02
Cost build-up and margin03
Proposal both engineering and sales can sign04
The full quoting workflow, connected.
Seven stages. Each one keeps the evidence it was handed.
See how Nyotta works01
Ingest
Drawing sets, STEP and native CAD, specifications, commercial workbooks, and the clarification thread.
02
Extract
Requirements, features, quantities, GD&T callouts, and the commercial terms that move the number.
03
Resolve
Rev conflicts, variant logic, missing inputs, and the assumptions nobody wrote down.
04
Check feasibility
Can we actually make it? Every route checked against your own machining, forging, and build criteria before a number is committed.
05
Estimate
Should-cost, tooling, secondary operations, and commercial impact, editable line by line.
06
Approve
The responsible engineer signs the output. Not the model.
07
Learn
Approved decisions carry into the next program with their evidence attached.
Built for work where the details are the job.
Four properties that decide whether a quoting system survives contact with a real package.
Every approval is context the next quote starts from.
An approved quote is not a number. It is a set of decisions, mappings, standards, and dispositions that already passed review. Nyotta keeps them with their evidence and the engineer's name attached, so the tenth program in a customer's book opens against your basis instead of a blank model.
All of it stays scoped to your instance. Nothing you approve trains a shared model or reaches another customer.
How isolation worksConnected context
What your systems already hold. Drawing sets, specifications, workbooks, closed quotes, rate cards, and the clarification threads, read where they live.
Learned context
What your engineers decide. A disposition on an exception, a corrected forge allowance, a customer-specific standard read one particular way. Each becomes a rule with its source, its owner, and the date it was approved, applied to the next package.
Judgment
Still yours. The layer below narrows what an engineer has to re-decide. It never decides for them, and nothing reaches the ERP without a signature.
We build the connection. You don't stage the data.
We build and maintain the connectors. Read-only to start, no migration, no data prep, and your quote history comes with it.
PLM & CAD
Drawings, models, revisions, lifecycle stateERP & procurement
Materials, suppliers, rate cards, cost historyCRM
Accounts, opportunities, quote outcomesWorkspace & knowledge
Specifications, standards, decision records
Drawings, models, revisions, lifecycle stateERP & procurementMaterials, suppliers, rate cards, cost historyCRMAccounts, opportunities, quote outcomesWorkspace & knowledgeSpecifications, standards, decision recordsSee every connector
Not on the list? Tell us what you run
Your engineering context stays controlled.
Scoped access, tenant separation, source-level traceability, and an engineer's approval before anything is written back.
Explore security01SOC 2 Type II
02Encryption in transit and at rest
03No cross-tenant training on customer data
04Approval before write-back
Operators who have worked both sides of the problem.
Hardware programs and production AI infrastructure on one side; language systems and enterprise search on the other. Nyotta is what we wanted to exist while doing that work.
Maddie Zheng
Co-founder & CEOLed the Pixel A-series program at Google from concept to launch. Built AI infrastructure for large-scale models at Meta Superintelligence Labs, including trust and safety systems running inline at production scale. Dropped out of Harvard Business School to build Nyotta AI.
Yiyuan Zhang
Co-founder & CTOLed NLP and AI projects at Google Assistant, now part of Google DeepMind. Built enterprise search at Moveworks for Northrop Grumman, CVS, and other enterprises with the most fragmented legacy systems and highest security requirements. Graduate degree in CS and NLP from Yale.
Common questions about Nyotta.
What does Nyotta AI do?
Nyotta reads engineering RFQ packages and turns them into a quote your team can review. It takes the whole package: 2D drawings, 3D CAD, specifications, variant tables, bid forms and the emails that clarify scope. Out of it comes a quote in your own template, ready to send: [including] a structured bill of materials, a should-cost estimate, a feasibility check, an open exception list and the outgoing proposal etc. Every line links back to the sheet and zone it came from, with the reasoning behind it shown.
Who is Nyotta for?
Nyotta is built for teams whose quoting is technical and high-variant: every RFQ brings a different set of parts, options and requirements, and the number that goes out has to be defended line by line. Estimating, applications and design engineering, sales, finance and purchasing all work from the same package rather than separate copies of it. For example, automotive Tier-1 suppliers quote engineered components from customer prints, and custom automation builders scope complete systems.
Which teams use Nyotta?
Estimating and quoting run it first, because the RFQ lands with them. Applications and design engineering work from the extracted requirements and drawing data instead of rebuilding context per bid. Sales gets a defensible number early enough to matter. Finance and purchasing see the cost drivers behind an approval rather than a standard cost. Everyone reads the same package, with the same source links.
Can Nyotta read a 2D mechanical drawing?
Yes. Nyotta reads 2D engineering drawings directly and extracts dimensions, geometric tolerancing callouts, threads, surface finishes, material and heat treat requirements, title block data and drawing notes. It also works on vector PDFs and on scanned prints with no text layer. Each extracted characteristic carries its location on the sheet, so the output can be checked against the drawing.
Can Nyotta read 3D CAD models?
Yes. Nyotta reads a model for overall envelope, true material volume and mass, wall thickness, and how much stock has to come off. Those feed the cost side, because mass and machined volume are among the strongest drivers of what a part actually costs. Send a drawing package, send a model, or send both, and Nyotta reads whichever arrives.
How long does it take to turn an RFQ package into a quote?
Minutes for the whole package, seconds for a sheet. A complex RFQ package of drawings, specifications and commercial documents is read end to end in minutes. It is read in full while the clarification window is still open, so questions reach your customer in time to be answered. What remains is engineering review, the part of quoting that needs a person.
Can we quote more RFQs without adding headcount?
Yes. That is the return most teams are buying. What goes away is the manual reading: working through forty-page specification PDFs, pulling features off drawings by hand, redrawing a customer print to generate a parts list, retyping tables into a cost sheet, chasing the same clarification twice. That work is measured in weeks per programme, and programme volume only goes up. Engineers still approve every quote before it goes out, and review is where their judgment was always the point.
How accurate is Nyotta at reading engineering drawings?
Nyotta scores 0.91 F1 on engineering drawing extraction. A general-purpose AI agent scores 0.38 F1 on the same work. Scored separately against released feasibility analyses, Nyotta reached 93.7% precision at 97.2% recall. Accuracy is re-measured on every pipeline change against golden sets built from real drawings, and each engagement runs to acceptance criteria both parties define before work starts.
What should we bring to a demo?
One RFQ package your current process struggles with. The drawings, the specification, the variant table or quantity breaks, and if you have it the quote you eventually sent. Nyotta runs it live and you compare the output against what your own team produced. The most useful demos have been on packages the customer already quoted, because they already know the right answer.
How long until we see results on our own data?
Quickly, because there is no data preparation step. The estimating templates, rate card, past proposals and historical RFQ packages already exist inside your business and are sent as they are, with no cleanup. You run your own programs through the platform from the start, and the pace after that is set by your review cycle rather than by anything Nyotta has to build first.