Report · Chapter 1

Project introduction and data source

How a public U.S. Army Corps of Engineers RFI register becomes the foundation of a Digital Engineering analytics and AI platform.

Digital Engineering RFI Analytics & AI Platform 3 October 2026
A tower crane over a concrete building frame at sunset.

Section 1Project introduction

This project develops a Digital Engineering RFI Analytics & AI Platform using a real procurement-stage Request for Information (RFI) / bidder-clarification dataset from a U.S. Army Corps of Engineers infrastructure procurement.

The goal is to show how project information stored in spreadsheets and other semi-structured sources can become a structured digital workflow for project reporting, engineering coordination, document control, automation, and AI-assisted analysis.

Illustrative upstream project workflow and RFI analytics pipeline

Scope and provenance. The actual dataset is a procurement-stage RFI / bidder-clarification register released publicly by the U.S. Army Corps of Engineers (USACE) with solicitation W912DS26BA032. The original internal system used to collect, review, and compile the questions is not verified. Autodesk Forma is presented below solely as an illustrative CDE / project-systems architecture, not as the proven origin of the Excel workbook.

For the illustrative upstream process, a project team could configure Forma Data Management as its common data environment (CDE), define information requirements, and support drawings, specifications, and models created by engineers using tools such as Revit, Civil 3D, or AutoCAD. Document-control or EDMS capabilities operate throughout information development, review, authorisation, issue, and archival, rather than appearing as one step after publication.

  • Common data environment (CDE). A governed approach to collecting, managing, and exchanging project information. In the example architecture, Forma Data Management provides the controlled files, access, versioning, and review environment.
  • Electronic document management (EDMS). The document-control functionality supporting registration, metadata, revisions, reviews, permissions, transmittals, and audit records across the lifecycle. A CDE may incorporate these EDMS capabilities.
  • Project Systems Engineer (illustrative role). Supports configuration, permission controls, workflow testing, integration, troubleshooting, and analytics; design authorisation and technical answers remain with the responsible engineering and procurement personnel.

ISO 19650 commonly describes four CDE information states, not necessarily four physical folders:

  1. Work in progress. Information being developed and checked by its responsible task team.
  2. Shared. Information made available for coordination or review under agreed controls.
  3. Published. Information authorised for a specified purpose; here, the relevant example is issued-for-tender documentation, not automatically issued-for-construction documentation.
  4. Archived. The retained history and record of superseded or exchanged information.

After procurement drawings and specifications are released, prospective bidders can raise technical clarification questions. The contracting and engineering teams register, allocate, investigate, and respond to them. In a hypothetical implementation those questions might be tracked in a digital workflow system, but the source does not establish that USACE used Forma Build. Nor does it establish that every RFI was closed before the register was published.

How to read the workflow: Steps −7 to −1 describe an illustrative upstream information-management and procurement sequence. Step 0 identifies the verified public Excel source. From Step 01 onward, this portfolio project ingests and analyses that workbook. Steps 05–07 describe planned advanced analytics and AI capabilities, not established features of the original source system.

  1. Project requirements and CDE setupIllustrative: define information requirements, naming, metadata, permissions, user groups, and controlled document workflows in a platform such as Forma Data Management.
  2. Site planning and initial designIllustrative: planners and engineering disciplines establish site context and start preparing design information; Forma Site Design could be used where relevant.
  3. Design work in progressIllustrative: engineering / BIM teams develop models, drawings, and technical specifications using authoring tools; document control and revision history apply throughout.
  4. Shared for coordination and reviewIllustrative: responsible teams share selected information for checking and coordination; project systems support access, review routing, and traceability.
  5. Documents authorised for tenderIllustrative: the authorised tender drawings and specifications are issued for procurement under agreed document-control procedures, not automatically for construction.
  6. Procurement package publishedUSACE releases solicitation documents for prospective bidders. A project CDE/EDMS could support preparation and issue, but the actual internal platform is unverified.
  7. Tender RFIs registered and answeredBidders ask clarification questions; procurement and technical staff allocate and respond. Forma Build is only an illustrative possible workflow, not a verified USACE source.
  8. Public RFI register published as ExcelVerified source: “RFI log for posting 9.23.26.xlsx” is made available with the procurement opportunity. Its original internal software and export method are unknown.
  9. Raw RFI Excel dataThis analytics platform starts here: preserve the public workbook as data/raw/rfi_log.xlsx and maintain source lineage.
  10. Python cleaning and validationIngest the raw source, standardise records, test quality, and retain raw-versus-derived distinctions.
  11. Structured SQL databaseOrganise RFI records and validated analytical attributes into a queryable reporting model.
  12. Power BI reportingExplore RFI volume, response coverage, disciplines, document impacts, and detailed records, within the data's limitations.
  13. Databricks analytics (planned)Proposed extension for more advanced data analysis and experiment tracking.
  14. Machine learning and NLP models (planned)Proposed classification and language-analysis experiments using clearly labelled derived attributes.
  15. Semantic search and RFI AI Copilot (planned)Proposed retrieval and question-answering layer requiring explicit source grounding and evaluation.

The work is designed to demonstrate capability across:

  • Digital Engineering
  • RFI workflow management
  • Project systems
  • Engineering information management
  • Python automation
  • SQL and data modelling
  • Power BI
  • Databricks
  • Machine learning
  • Natural language processing
  • Generative AI
  • AI evaluation and governance

A major strength of the project is that it begins with a real publicly released procurement RFI register, rather than a synthetic portfolio dataset.

Section 2Business context

Infrastructure projects generate clarification requests during procurement and RFIs during design and construction delivery. This source specifically documents procurement-stage bidder questions and official responses.

An engineer on a construction site reviewing a building model, charts, and drawings on a tablet.
Drawings, schedules, and site records reviewed together.

RFIs are used when project participants need clarification on issues such as:

  • unclear drawings
  • conflicting specifications
  • missing information
  • constructability questions
  • scope gaps
  • coordination problems
  • compliance questions
  • material or equipment requirements

Because much of this information is stored as free text, it can be difficult to identify recurring engineering issues, document impacts, technical themes, or common clarification categories.

This project turns the RFI register into a structured analytical and AI-enabled platform.

Section 3Data source

The dataset comes from the U.S. Army Corps of Engineers solicitation Power Generation with Microgrid at Joint Base McGuire-Dix-Lakehurst (JBMDL), New Jersey.

Solicitation referenceW912DS26BA032

The project involves power-generation and microgrid infrastructure at Joint Base McGuire-Dix-Lakehurst in New Jersey. The wider technical context includes:

  • electrical generation
  • battery energy storage
  • solar photovoltaic systems
  • SCADA and control systems
  • cybersecurity
  • civil works
  • mechanical and utility systems
  • specifications
  • design drawings
  • inspection requirements

Section 4Source file

The verified source used in this project is the Excel workbook published with the federal procurement opportunity. Its internal authoring, document-management, and export software are not identified; Autodesk Forma is not established as its source.

WorkbookRFI log for posting 9.23.26.xlsx

The workbook contains approximately 111 RFI records. The original workbook will be preserved unchanged as the raw project source.

Section 5Explore the dataset

Official SAM.gov opportunity

The official federal procurement opportunity is published on SAM.gov. In the Attachments section, look for RFI log for posting 9.23.26.xlsx.

Public dataset preview

A public indexed copy of the RFI workbook can also be explored on GovTribe.

Section 6Original dataset structure

The source workbook contains these main fields:

  • RFI #
  • FROM
  • TASKED
  • DATE RECEIVED
  • QUESTION
  • RESPONSE
  • NOTES
  • Status

RFI #

Identifier assigned to each Request for Information.

FROM

Identifies the organisation or party that submitted the RFI. This field supports counts by submitting organisation, topics by organisation, and the most active submitting parties.

TASKED

Identifies the party or engineering function responsible for reviewing or responding to the RFI.

DATE RECEIVED

The date on which the RFI was received. This supports RFI volume over time, weekly trends, and peak clarification periods.

QUESTION

The original engineering or technical question. This is one of the most valuable fields because it contains unstructured engineering text. Potential derived attributes include discipline, issue type, technical topic, drawing reference, specification reference, equipment reference, and a question summary.

RESPONSE

The official answer or clarification associated with the RFI. This supports question-response analysis, technical resolution analysis, summarisation, and semantic search.

NOTES

Supplementary information that can help identify whether an RFI caused a drawing update, a specification update, an attachment issue, a document revision, or no document change.

Status

The source status where it is populated. The original source status will be preserved. If required, a separate derived status may be created, such as Answered, Open, or Unknown.

Section 7Why this dataset is suitable

The dataset is suitable because it combines a real construction context, structured project fields, unstructured engineering text, genuine RFI questions and responses, and relevant Digital Engineering and document-control use cases.

The technical content supports classification across:

  • Electrical
  • Mechanical
  • Civil
  • Structural
  • Controls / SCADA
  • Cybersecurity
  • Quality / Inspection
  • Specifications
  • Design coordination

Section 8Planned transformation

The project will maintain clear lineage between original source data and derived fields.

  1. Original Excel
  2. raw_rfi
  3. clean_rfi
  4. enriched_rfi
  5. SQL and Power BI
  6. Databricks machine learning and AI

Raw layer

The raw layer keeps the ingested record close to the source:

  • rfi_number
  • from_party
  • tasked_to
  • date_received
  • question
  • response
  • notes
  • status_raw
  • source_file
  • ingestion_timestamp

Clean layer

The clean layer standardises RFI identifiers, party names, dates, whitespace, null values, text encoding, status, and duplicates.

Enriched layer

The enriched layer adds derived analytical and AI attributes:

  • discipline
  • issue_type
  • technical_topic
  • reference_type
  • reference_number
  • drawing_update_required
  • spec_update_required
  • document_update_required
  • question_summary
  • response_available
  • classification_confidence

Section 9Data limitations

The source contains approximately 111 RFI records, which is relatively small for supervised machine learning.

It also does not directly contain all fields expected in a mature RFI system, such as due date, response date, closed date, formal priority, cost impact, schedule impact, discipline, issue category, workflow history, and revision history.

Any additional fields derived using rules, manual labelling, or AI will be clearly marked as derived attributes. The project will not present inferred values as if they were present in the original source.

Section 10Planned use of the dataset

Python and SQL

  • Data ingestion
  • Data cleaning
  • Validation
  • Transformation
  • Data modelling

Power BI

  • RFI overview
  • RFI trend analysis
  • RFI by discipline
  • RFI by issue type
  • Document impact analysis
  • Detailed RFI register

Databricks and machine learning

  • Discipline classification
  • Issue-type classification
  • Document-impact prediction
  • MLflow experiment tracking
  • Model Registry

Generative AI

  • RFI summarisation
  • Semantic search
  • Retrieval-augmented generation
  • Natural-language RFI querying
  • RFI AI Copilot
  • AI evaluation

Section 11Project value

The final system demonstrates how a traditional RFI register can be converted into a modern Digital Engineering information platform.

A real USACE RFI workbook is ingested and cleaned using Python, modelled in SQL, visualised through Power BI, and extended in Databricks with NLP classification, document-impact prediction, semantic search, and an AI Copilot for engineering information retrieval.
  • Construction
  • Digital Engineering
  • Project systems
  • Data analytics
  • Automation
  • AI