Contacts
Get in touch
Close

Contacts

Romania, Cluj-Napoca – 400157
Str. Ploiești Nr. 9

Romania, Vama – 447350
Str. Fermei Nr. 14

+40 774 463 429

info@inteliscience.net

AI Similar Project Search

AI Similar Project Search

AI Similar Project Search: From a New RFQ to the Projects You Have Already Built

An AI similar project search reads a new RFQ package and shows which completed projects are technically closest, why each one matched and where the evidence sits. Your engineers start the quotation from a project the company has already built, not from a blank page and not from whoever happens to remember one.

Short answer

Similar project search compares a new RFQ with completed engineer-to-order projects to find those that already solved a comparable technical problem. With AI, each hit shows why it matched, its documents and the source passages.

inteliScience builds this read-only next to your PDM, ERP, wiki and file server.

This page belongs to Engineering Data AI and covers the project level: a whole RFQ against whole past projects. Comparing parts by their shape, to find the nearest existing component or a duplicate, is a different job with a different index, covered in CAD similarity search. How AI reads a specification in the first place is explained in AI requirements specification analysis.

Why finding similar past projects takes so long

Handling information is a measurable share of engineering work. In a work-sampling study of 78 design engineers over 20 working days, information behaviours, from looking for information to receiving and providing it, accounted for 55.75% of working time (Robinson, JASIST 2010). The study is not about quotations, but it shows how much engineering time goes into finding and passing on information. Research on engineer-to-order quotation points the same way: decision support that gives access to detailed and accurate information during quotation preparation enables a high level of product adaptation while keeping the company efficient (Elgh, Advanced Engineering Informatics, 2012).

What AI similar project search shows for each hit

Similar projects

Which past projects are close

The completed projects that are technically similar to the new RFQ, found by their requirements and features rather than by name.

Reason

Why each one matched

The reason a project was rated similar, stated for every hit, so the match can be checked instead of trusted.

Documents

What exists for the project

Which documents and information are on record for it, such as the specification, the quotation, reports and notes.

Citations

The source passages

The direct passages in those documents that the match rests on, so the engineer reads the original wording.

PDM

The project’s PDM records

Part and assembly numbers, names, structure and project assignment, linked to the hit.

ERP

Article, cost and procurement data

The ERP information behind the project, read-only, as the starting point for a new estimate.

Assembly and test station for a drive unit

Illustrative hit · invented data
Why similar
Same joining process, a comparable cycle-time requirement, and the same demand for a documented test result per unit.
Documents
Requirements specification, quotation, acceptance report, lessons-learned page in the wiki.
Source passage
Specification, testing section: Each unit shall be tested and the result stored with its serial number.
PDM
Top-level assembly of the station with its structure and project assignment.
ERP
Article, cost and procurement information for the station and its main purchased components.

An illustration of the structure of a hit, not data from any real project.

How a similar project search runs, from RFQ package to past projects

  1. Hand over the RFQ package. Requirements specification, technical PDFs, drawings and further requirements, in the form the customer sent them.
  2. Recognise the requirements. The AI identifies the technical requirements and features that matter for the comparison, for example performance data, dimensions, materials, ambient conditions, applicable standards and interfaces.
  3. Search the completed projects. Across project documents on the file server, pages in the wiki and the project metadata from the PDM: part and assembly numbers, names, structure and project assignment.
  4. Explain each match. For every similar project, the reason it was rated similar.
  5. Show the evidence. The documents that exist for the project and the direct source passages, so the engineer can verify the match in the original.
  6. Link to PDM and ERP. Each hit leads to its PDM records and to the ERP article, cost and procurement data, read-only.

Which systems the search reads from

Systems the project search reads from, with examples, the read-only access route and what each adds to a hit
System Examples How it is read What it adds to a hit
PDM A commercial PDM, or a PDM built in-house on a SQL database Read-only; for a PDM on a SQL database, direct read access is unproblematic, other routes are clarified in the Quick Scan Part and assembly numbers, names, structure and project assignment
ERP SAP S/4HANA, SAP Business One, proAlpha or Microsoft Dynamics 365 Read-only through the vendor’s standard interfaces, for example the SAP S/4HANA OData services, the SAP Business One Service Layer or the Business Central API Article, cost and procurement information
Wiki Confluence, SharePoint or a comparable intranet Standard connectors, no special restrictions Decisions, lessons learned and project notes
File server Windows file shares or document libraries Standard connectors, no special restrictions Specifications, quotations, reports and technical PDFs
CAD Siemens NX, CATIA, SolidWorks or Creo Not needed for the metadata-based project search Geometry only matters for comparing parts by shape, which needs its own index

What similar project search does not do

Good fit

  • Engineer-to-order, special machinery and plant engineering with recurring but never identical requests
  • Requirements specifications and technical PDFs as the normal form of an inquiry
  • Completed projects documented across PDM, ERP, wiki and file server
  • Teams in which finding the right reference project depends on a few experienced people

Limits we state up front

  • No writing into PDM, ERP, wiki or file server, and no replacement of any of them
  • No comparison of part geometry without a geometry index
  • No automatic quotation price: the search shows past cost data, the estimate stays with your calculation
  • No bid or no-bid decision; the engineer chooses the reference project
  • No silent cleanup of master data: if the project number does not run through, mapping it is visible project work

Deployment, data and hosting

Where the search runs, which language models it uses and which users may see which projects and cost data are settled with your IT before the proof of value starts. How a private retrieval assistant runs entirely inside a company’s own infrastructure is described in our private AI knowledge assistant case study.

How we introduce AI project search

  1. Quick Scan with on-site workshop, one week. Up to two workshop days, five days in total: which systems hold project knowledge, how they can be read, and whether a project or order number runs consistently through PDM, ERP, wiki and file server.
  2. Scope the proof of value. One product family and a handful of completed projects, chosen with your engineers.
  3. Connect read-only. PDM metadata, ERP article, cost and procurement data, and the wiki pages and file server documents of those projects.
  4. Test with RFQs you have already answered. An RFQ whose reference projects your engineers know is the fairest test: does the search find them, and are the stated reasons right?
  5. Decide on the next step. More product families, more systems and, where shape matters, a geometry index for CAD similarity search. We can take overall responsibility for both from one hand.

Bounded scope · on your data

NX Sprint

One product family and a handful of completed projects: a running search on your data, and an honest picture of your data quality.

  • Preceded by a Quick Scan with an on-site workshop: systems, access, project numbers
  • Read-only connection to PDM, ERP, wiki and file server for the chosen projects
  • Hits with reasons, documents, source passages and PDM and ERP links
  • A clear statement of what a rollout needs, driven by your data rather than by a slide
Discuss an NX Sprint

Read-only integration · your systems stay in place

For comparison

Searching by hand

The reference project found by memory, folder by folder and system by system.

  • Search by number in the PDM, by keyword on the file server
  • Costs looked up separately in the ERP
  • The result depends on who is in the office that week

Frequently asked questions about similar project search

What is similar project search in engineering?
Similar project search is the comparison of a new customer request, usually a requirements specification or RFQ package, with completed projects to find the ones that solved a comparable technical problem. In engineer-to-order companies it decides whether a quotation starts from proven work or from scratch. With AI, the comparison runs on technical requirements and features rather than on project names or numbers.
How can AI find similar past projects from a new RFQ?
AI finds similar past projects by recognising the technical requirements and features in the RFQ package and comparing them with the documents and metadata of completed projects. In our implementation each hit shows why it was rated similar, which documents exist for it and the direct source passages, and links to the PDM records and to ERP article, cost and procurement data.
Why does data quality matter for finding similar past projects?
Data quality matters because a similar project is only useful if its documents, PDM records and ERP costs can be joined. The deciding factor is whether a project or order number runs consistently through PDM, ERP, wiki and file server. If it does, linking is simple; if it does not, it is solvable but becomes part of the project, which is why we look at it first in a Quick Scan with an on-site workshop.
What is the difference between similar project search and CAD similarity search?
Similar project search compares a whole RFQ with whole past projects using requirements, documents and metadata; CAD similarity search compares parts by their geometry. Geometry search needs a one-time index over the part library, whether it comes from a specialised tool such as CADENAS or Siemens Geolus or is built for your system directly. Pure geometry tools do not cover specification analysis, document search with sources or links to project cost data. The two complement each other, and we can take overall responsibility for both from one hand.
How much effort does it take to introduce AI similar project search in an engineer-to-order company?
The effort depends mainly on the quality and linkability of the existing data, not on the AI. A sensible start is a bounded proof of concept, such as our NX Sprint, on one product family and a handful of completed projects, preceded by a Quick Scan with an on-site workshop that looks at systems, access and project numbers. The result is a running search on your data and a realistic basis for planning the rollout.

Name the last RFQ your team quoted from memory.

Describe the kind of request, not the documents. We will tell you which systems the search would read, what your project numbers mean for the effort, and what a proof of value on one product family would show.

Talk to an Engineer

Read-only integration · whatever CAD you run · inteliScience SRL, nearshore from Romania

Cookie settings

Choose which services we may use on this website. Your choice is stored in your browser for 12 months. You can change it at any time via “Cookie settings” at the bottom of every page.

Strictly necessary

Stores your cookie choice. This cookie cannot be switched off.

Always active
Show cookies
NameProviderPurposeStorage period
is_cookie_consentinteliScienceStores your cookie choice and when you made it12 months

Google Analytics 4 measures page views, traffic sources and form use with a pseudonymous ID. Provider: Google Ireland Ltd. Data may also be processed by Google LLC in the USA (EU–US Data Privacy Framework).

Show cookies
NameProviderPurposeStorage period
_gaGoogleDistinguishes visitors with a pseudonymous ID2 years
_ga_<ID>GoogleStores the session state2 years

Google Ads measures which ads lead to enquiries (conversions) and lets us show our ads to previous visitors (remarketing). Provider: Google Ireland Ltd. Data may also be processed by Google LLC in the USA (EU–US Data Privacy Framework).

Show cookies
NameProviderPurposeStorage period
_gcl_au, _gcl_aw, _gcl_gsGoogleLinks ad clicks to enquiries90 days
_gcl_ls (local storage)GoogleLinks ad clicks to enquiries90 days
IDE (doubleclick.net)GoogleRemarketing and ad measurement13 months

Videos: YouTube videos only load when you press play (via youtube-nocookie.com). Only then does Google receive data.