AI-Driven Automation for Construction Drawing Upload
Co-Founder & Design Lead · 2019 · 1 Eng Lead, 2 Designers, 2 Engineers
AI-Driven Automation for Construction Drawing Upload
Co-Founder & Design Lead · 2019 · 1 Eng Lead, 2 Designers, 2 Engineers
AI-Driven Automation for Construction Drawing Upload
Co-Founder & Design Lead · 2019 · 1 Eng Lead, 2 Designers, 2 Engineers
TL;DR
I transformed a two-day, error-prone process into a seamless 10-minute workflow—a 100x improvement—by while balancing technical constraints and user needs.
TL;DR
I transformed a two-day, error-prone process into a seamless 10-minute workflow—a 100x improvement—by while balancing technical constraints and user needs.
Step 1: Splitting sheets
I worked closely with the engineering team to figure out the feasibility of automatically splitting the PDF, during upload and within the browser.Step 1.2: Data Extraction Through Template (Only first time)
The next step was naming the sheets and adding additional metadata to it. From my research, I knew all this data was present in the "Title Block".Step 2: Review Drawings
This step ensures accuracy by enabling human review of OCR results before upload - maintaining accuracy and control.Step 1: Splitting sheets
I worked closely with the engineering team to figure out the feasibility of automatically splitting the PDF, during upload and within the browser.Step 1.2: Data Extraction Through Template (Only first time)
The next step was naming the sheets and adding additional metadata to it. From my research, I knew all this data was present in the "Title Block".Step 2: Review Drawings
This step ensures accuracy by enabling human review of OCR results before upload - maintaining accuracy and control.Step 1: Splitting sheets
I worked closely with the engineering team to figure out the feasibility of automatically splitting the PDF, during upload and within the browser.Step 1.2: Data Extraction Through Template (Only first time)
The next step was naming the sheets and adding additional metadata to it. From my research, I knew all this data was present in the "Title Block".Step 2: Review Drawings
This step ensures accuracy by enabling human review of OCR results before upload - maintaining accuracy and control.Step 1: Splitting sheets
I worked closely with the engineering team to figure out the feasibility of automatically splitting the PDF, during upload and within the browser.Step 1.2: Data Extraction Through Template (Only first time)
The next step was naming the sheets and adding additional metadata to it. From my research, I knew all this data was present in the "Title Block".Step 2: Review Drawings
This step ensures accuracy by enabling human review of OCR results before upload - maintaining accuracy and control.
Step 1: Splitting sheets
I worked closely with the engineering team to figure out the feasibility of automatically splitting the PDF, during upload and within the browser.Step 1.2: Data Extraction Through Template (Only first time)
The next step was naming the sheets and adding additional metadata to it. From my research, I knew all this data was present in the "Title Block".Step 2: Review Drawings
This step ensures accuracy by enabling human review of OCR results before upload - maintaining accuracy and control.Step 1: Splitting sheets
I worked closely with the engineering team to figure out the feasibility of automatically splitting the PDF, during upload and within the browser.Step 1.2: Data Extraction Through Template (Only first time)
The next step was naming the sheets and adding additional metadata to it. From my research, I knew all this data was present in the "Title Block".Step 2: Review Drawings
This step ensures accuracy by enabling human review of OCR results before upload - maintaining accuracy and control.Step 1: Splitting sheets
I worked closely with the engineering team to figure out the feasibility of automatically splitting the PDF, during upload and within the browser.Step 1.2: Data Extraction Through Template (Only first time)
The next step was naming the sheets and adding additional metadata to it. From my research, I knew all this data was present in the "Title Block".Step 2: Review Drawings
This step ensures accuracy by enabling human review of OCR results before upload - maintaining accuracy and control.Step 1: Splitting sheets
I worked closely with the engineering team to figure out the feasibility of automatically splitting the PDF, during upload and within the browser.Step 1.2: Data Extraction Through Template (Only first time)
The next step was naming the sheets and adding additional metadata to it. From my research, I knew all this data was present in the "Title Block".Step 2: Review Drawings
This step ensures accuracy by enabling human review of OCR results before upload - maintaining accuracy and control.
Step 1: Splitting sheets
I worked closely with the engineering team to figure out the feasibility of automatically splitting the PDF, during upload and within the browser.Step 1.2: Data Extraction Through Template
The next step was naming the sheets and adding additional metadata to it. From my research, I knew all this data was present in the "Title Block".Step 2: Review Drawings
This step ensures accuracy by enabling human review of OCR results before upload - maintaining accuracy and control.Step 1: Splitting sheets
I worked closely with the engineering team to figure out the feasibility of automatically splitting the PDF, during upload and within the browser.Step 1.2: Data Extraction Through Template
The next step was naming the sheets and adding additional metadata to it. From my research, I knew all this data was present in the "Title Block".Step 2: Review Drawings
This step ensures accuracy by enabling human review of OCR results before upload - maintaining accuracy and control.Step 1: Splitting sheets
I worked closely with the engineering team to figure out the feasibility of automatically splitting the PDF, during upload and within the browser.Step 1.2: Data Extraction Through Template
The next step was naming the sheets and adding additional metadata to it. From my research, I knew all this data was present in the "Title Block".Step 2: Review Drawings
This step ensures accuracy by enabling human review of OCR results before upload - maintaining accuracy and control.Step 1: Splitting sheets
I worked closely with the engineering team to figure out the feasibility of automatically splitting the PDF, during upload and within the browser.Step 1.2: Data Extraction Through Template
The next step was naming the sheets and adding additional metadata to it. From my research, I knew all this data was present in the "Title Block".Step 2: Review Drawings
This step ensures accuracy by enabling human review of OCR results before upload - maintaining accuracy and control.
Context
Construction projects involve 1000s of drawings; shared in batches—both digitally via the cloud and as physical copies—across all stakeholders.
Drawings or Blueprints serve as the primary tool for coordination, execution and approvals among architects, engineers, contractors, and owners. But with so many versions flying around, staying aligned is a constant challenge.
Context
Construction projects involve 1000s of drawings; shared in batches—both digitally via the cloud and as physical copies—across all stakeholders.
Drawings or Blueprints serve as the primary tool for coordination, execution and approvals among architects, engineers, contractors, and owners. But with so many versions flying around, staying aligned is a constant challenge.
Business Goals
To grow Buildsys, we needed to prove ROI and drive adoption—drawings were our highest-leverage entry point.
Rework – fixing errors caused by outdated or miscommunicated drawings – is one of the biggest drains on time and budget in construction. We focused our first module here with two key business goals:
Business Goals
To grow Buildsys, we needed to prove ROI and drive adoption—drawings were our highest-leverage entry point.
Rework – fixing errors caused by outdated or miscommunicated drawings – is one of the biggest drains on time and budget in construction. We focused our first module here with two key business goals:
Challenge
Managing drawings on the cloud is a tedious and painful manual process, with a lot of room for mistakes. It takes 2 days to upload a set of 100 drawings!
This process is so high stakes that companies hire a trained engineer solely for drawing management.
Challenge
Managing drawings on the cloud is a tedious and painful manual process, with a lot of room for mistakes. It takes 2 days to upload a set of 100 drawings!
This process is so high stakes that companies hire a trained engineer solely for drawing management.
Research
43 participants across seven different stakeholders and two cities.
To understand the context, I along with my design team conducted 1:1 interviews, secondary research and field visits, visiting different types of people involved in construction.
Research
43 participants across seven different stakeholders and two cities.
To understand the context, I along with my design team conducted 1:1 interviews, secondary research and field visits, visiting different types of people involved in construction.
What we observed
Drawings are distributed at different stages as packages often with 100+ sheets
Drawings are of many different kinds or trades and are distributed at different stages of the project as packages. Each package often contains with 100+ sheets.
Snippet from a planning document
Each drawing has 9+ classification attributes - leading to complex folder structures and naming conventions.
Each drawing has many classification attributes such as project name, stage, trade, company name etc.
All this info is present inside the title block of the drawing
Typical title block: All important drawing metadata is exists in the title block
Drawings undergo multiple revisions at every stage of the project.
People employ different strategies to handle revisions such as adding timestamps, archiving folders or deleting files.
Dropbox showing different revisions categorized using timestamps
What we observed
Drawings are distributed at different stages as packages often with 100+ sheets
Drawings are of many different kinds or trades and are distributed at different stages of the project as packages. Each package often contains with 100+ sheets.
Snippet from a planning document
Each drawing has 9+ classification attributes - leading to complex folder structures and naming conventions.
Each drawing has many classification attributes such as project name, stage, trade, company name etc. All this info is present inside the title block of the drawing.
Typical title block: All important drawing metadata is exists in the title block
Drawings undergo multiple revisions at every stage of the project.
People employ different strategies to handle revisions such as adding timestamps, archiving folders or deleting files.
Dropbox showing different revisions categorized using timestamps
What we observed
Drawings are distributed at different stages as packages often with 100+ sheets
Drawings are of many different kinds or trades and are distributed at different stages of the project as packages. Each package often contains with 100+ sheets.
Snippet from a planning document
Each drawing has 9+ classification attributes - leading to complex folder structures and naming conventions.
Each drawing has many classification attributes such as project name, stage, trade, company name etc.
All this info is present inside the title block of the drawing
Typical title block: All important drawing metadata is exists in the title block
Drawings undergo multiple revisions at every stage of the project.
People employ different strategies to handle revisions such as adding timestamps, archiving folders or deleting files.
Dropbox showing different revisions categorized using timestamps
How might we
HMW make drawing uploads faster and less error prone without disrupting users’ mental models?
We mapped the users’ mental model and asked HMW questions at each step.
Mental Model
How might we
How might we
I have a PDF with 120 sheets that i need to split into 120 separate PDFs.
Take rough notes, then organize later
… reduce the manual overhead associated with splitting PDFs into individual sheets
I need to name each PDF based on stage, trade, scale, sheet size etc.
Work fast with keyboard
… reduce the amount of manual data entry required
... validate that the extracted info against the title blocks
... allow users to review and correct extracted info
I need to ensure all stakeholders are using the latest revision.
Forced a format, not flexible.
… ensure drawing revisions are handled correctly.
How might we
HMW make drawing uploads faster and less error prone without disrupting users’ mental models?
We mapped the users’ mental model and asked HMW questions at each step.
Mental Model
How might we
How might we
I have a PDF with 120 sheets that i need to split into 120 separate PDFs.
Take rough notes, then organize later
… reduce the manual overhead associated with splitting PDFs into individual sheets
I need to name each PDF based on stage, trade, scale, sheet size etc.
Work fast with keyboard
… reduce the amount of manual data entry required
... validate that the extracted info against the title blocks
... allow users to review and correct extracted info
I need to ensure all stakeholders are using the latest revision.
Forced a format, not flexible.
… ensure drawing revisions are handled correctly.
How might we
HMW make drawing uploads faster and less error prone without disrupting users’ mental models?
We mapped the users’ mental model and asked HMW questions at each step.
Mental Model
How might we
How might we
I have a PDF with 120 sheets that i need to split into 120 separate PDFs.
Take rough notes, then organize later
… reduce the manual overhead associated with splitting PDFs into individual sheets
I need to name each PDF based on stage, trade, scale, sheet size etc.
Work fast with keyboard
… reduce the amount of manual data entry required
... validate that the extracted info against the title blocks
... allow users to review and correct extracted info
I need to ensure all stakeholders are using the latest revision.
Forced a format, not flexible.
… ensure drawing revisions are handled correctly.
Hypothesis & Success Metrics
By automating PDF splitting, title block data extraction, and version control, we aim to:
Reduce upload time from ~2 days to under 10 minutes
Cut manual data entry by 80–90%
Improve metadata extraction accuracy to >95%
Reduce rework due to outdated sheets
Preserve user mental models and workflows
Hypothesis & Success Metrics
By automating PDF splitting, title block data extraction, and version control, we aim to:
Reduce upload time from ~2 days to under 10 minutes
Cut manual data entry by 80–90%
Improve metadata extraction accuracy to >95%
Reduce rework due to outdated sheets
Preserve user mental models and workflows
Key Decisions & Tradeoffs
I partnered with engineering to make research-backed decisions that balanced UX and feasibility.
Key Decisions & Tradeoffs
I partnered with engineering to make research-backed decisions that balanced UX and feasibility.
Concept Validation
Lo-fi user testing revealed hidden friction we couldn’t have caught in early research
In our first prototype, we included an extra step asking users to manually select trade and stage for each drawing.
Through testing with construction teams, we discovered this step was unnecessary and even confusing. The revision number already accounted for trade and stage in their existing workflows.
This insight helped us remove redundant inputs, streamline the upload flow, and align more closely with how users mentally organize drawings - something we wouldn’t have uncovered through interviews alone.
Concept Validation
Lo-fi user testing revealed hidden friction we couldn’t have caught in early research
In our first prototype, we included an extra step asking users to manually select trade and stage for each drawing.
Through testing with construction teams, we discovered this step was unnecessary and even confusing. The revision number already accounted for trade and stage in their existing workflows.
This insight helped us remove redundant inputs, streamline the upload flow, and align more closely with how users mentally organize drawings - something we wouldn’t have uncovered through interviews alone.
From… To…
Reducing hundreds of steps to just 3 steps
Old Flow
Split files manually → Rename → Organize folders → Manage versions.
New Flow
Upload → Extract → Review → Done
From… To…
Reducing hundreds of steps to just 3 steps
Old Flow
Split files manually → Rename → Organize folders → Manage versions.
New Flow
Upload → Extract → Review → Done
From… To…
Reducing hundreds of steps to just 3 steps
Old Flow
Split files manually → Rename → Organize folders → Manage versions.
New Flow
Upload → Extract → Review → Done
Outcome
Making upload process 100 times faster!
Each screen was designed to align with our key design principles: Guide through the new, Empower the user, and Honor the mental model.
Step 1: Sheet Splitting
Users start by uploading drawings in bulk from multiple sources—local device, Dropbox, Drive, or direct links.
Highlights
Familiar drag-and-drop pattern
Clean progress indicators
Clear feedback on sheet counts per PDF
Design Principle
Honor the mental model — we mirrored the structure users were already used to: packages → sheets.
Upload file from computer or cloud
Buildsys autosplits the PDF into sheets
Step 2: Data Extraction
The next step was naming the sheets and adding additional metadata to it. From my research, I knew all this data was present in the "Title Block". I collaborated with the engineering team to explore the feasibility of extracting this data using Optical Character Recognition (OCR).
Highlights
Smart reuse of previously created templates
Allows field tagging directly from drawings (OCR zones)
Design Principle
Empower, don't replace — users control what’s extracted, reinforcing trust in the automation.
Select or create a template to locate the title block
Define regions on the drawing sheet to extract metadata using OCR
Step 3: Revision Bumping, Review & Publish
Lastly, if the user uploaded a drawing sheet having an ID that already existed in our system, we automatically incremented the drawing revision number. The user reviews the extracted info for correctness and publishes the drawings.
Highlights
Inline editing, batch actions (like stage or discipline assignment)
Visual preview of title block for every sheet
Clear distinction of required vs optional fields
Design Principle
Guide through the new — Smart defaults, editable fields, and confidence messaging support every user action.
Review and correct extracted drawing information
Publish
Outcome
Making upload process 100 times faster!
Each screen was designed to align with our key design principles: Guide through the new, Empower the user, and Honor the mental model.
Step 1: Sheet Splitting
Users start by uploading drawings in bulk from multiple sources—local device, Dropbox, Drive, or direct links.
Highlights
Familiar drag-and-drop pattern
Clean progress indicators
Clear feedback on sheet counts per PDF
Design Principle
Honor the mental model — we mirrored the structure users were already used to: packages → sheets.
Upload file from computer or cloud
Buildsys autosplits the PDF into sheets
Step 2: Data Extraction
The next step was naming the sheets and adding additional metadata to it. From my research, I knew all this data was present in the "Title Block". I collaborated with the engineering team to explore the feasibility of extracting this data using Optical Character Recognition (OCR).
Highlights
Smart reuse of previously created templates
Allows field tagging directly from drawings (OCR zones)
Design Principle
Empower, don't replace — users control what’s extracted, reinforcing trust in the automation.
Select or create a template to locate the title block
Define regions on the drawing sheet to extract metadata using OCR
Step 3: Revision Bumping, Review & Publish
Lastly, if the user uploaded a drawing sheet having an ID that already existed in our system, we automatically incremented the drawing revision number. The user reviews the extracted info for correctness and publishes the drawings.
Highlights
Inline editing, batch actions (like stage or discipline assignment)
Visual preview of title block for every sheet
Clear distinction of required vs optional fields
Design Principle
Guide through the new — Smart defaults, editable fields, and confidence messaging support every user action.
Review and correct extracted drawing information
Publish
Impact
Drawing upload, along with drawings module as a whole, helped reduce staff costs by 20% and rework costs by almost 30%.
Drawing upload, along with other key innovations in the drawings module geared towards avoiding rework, helped
reduce staff costs by 20%,
rework costs by almost 30% and,
Buildsys go from pilot to essential daily tool.
Impact
Drawing upload, along with drawings module as a whole, helped reduce staff costs by 20% and rework costs by almost 30%.
Drawing upload, along with other key innovations in the drawings module geared towards avoiding rework, helped
reduce staff costs by 20%,
rework costs by almost 30% and,
Buildsys go from pilot to essential daily tool.
Learnings
User Likes, Business Loves
As designers, we strive to create the best user experience possible.
However, being a cofounder taught me to balance this with the realities of engineering time and cost, often necessitating trade-offs.
I worked closely with XFN to understand feasibility and prioritize features.
Learnings
User Likes, Business Loves
As designers, we strive to create the best user experience possible.
However, being a cofounder taught me to balance this with the realities of engineering time and cost, often necessitating trade-offs.
I worked closely with XFN to understand feasibility and prioritize features.