A construction site produces two buildings at once. One is concrete, steel, conduit and glass. The other is a shaky digital replica assembled from phone photos, progress reports, marked-up plans and messages that begin with “just checking in.” The physical building rises floor by floor. Its digital twin usually lands in a folder called Site Visit Final 3.
Ardalan Khosrowpour has spent much of his career trying to make that second building less ridiculous. His company, OnsiteIQ, sends people through active jobsites with 360-degree cameras, maps the resulting imagery to floor plans and repeats the ritual often enough to show what changed. The images can become a visual record for an owner in another city, a developer juggling several projects or an investor who would prefer evidence to optimism.
The camera is the easy part. The consequential idea is consistency. Capture the same site completely, on a dependable cadence, and an image stops being an anecdote. It gains a location, a date and a previous version. It can be compared. It can support a decision. Eventually, it can help train a machine to understand the progress of a building.
Construction was the family language
Khosrowpour has written that some of his earliest memories took place on his father's construction sites. He worked in the family business for a period, then studied civil and environmental engineering at Shiraz University. At the University of Illinois Urbana-Champaign, his 2013 master's thesis examined how depth images could help assess interior construction work. The title was academic; the instinct was practical. Could a camera reveal not merely what a place looked like, but what work was actually happening?
Halfway through that degree, he encountered an irritation that became a career. The construction principles in his coursework felt too similar to those his father had studied decades earlier. He began asking why technology had changed the administrative surface of the industry without changing enough of its underlying visibility. He shifted his focus toward artificial intelligence and went to Virginia Tech for a Ph.D. in civil engineering.
His doctoral research dealt with quantitative analysis in the built environment. Along the way came fellowships, an NSF I-Corps project, work as a machine-learning engineer at building-systems company Ecorithm and a visiting-scholar stint at ETH Zurich. By 2016 he had a particular collection of tools: construction experience, civil engineering, computer vision, machine learning and a growing suspicion that a poorly observed industry was paying for its blind spots.
“When I arrived at Cornell Tech, I knew I wanted to do something to digitize construction.”Ardalan Khosrowpour, 2018
A very late arrival
The hinge in the story was nearly missed in transit. Khosrowpour moved to New York and interviewed for Cornell Tech's Runway Startup Postdoc program in 2016. He recently recalled taking the wrong trains, getting caught in delays and arriving 45 minutes late. An entire committee had been waiting.
It is an excellent nightmare: a new city, an important interview and the subway politely rearranging your future. Khosrowpour entered stressed and embarrassed. Uzi de Haan and Shuli Cohen Shwartz did not sharpen the embarrassment. They let him settle, breathe and begin. Years later, remembering de Haan, Khosrowpour described that second chance as life-changing.
Runway gave technical founders a postdoctoral setting in which to turn research into companies. It also required the less romantic labor of customer discovery. Khosrowpour spoke with contractors, owners, insurers and other stakeholders. Jobsite cameras sounded obvious until privacy, constantly moving work and incomplete coverage made them useless. A fixed feed could see one patch of chaos very well. A walking 360-degree camera could see the building.
At Cornell Tech, Khosrowpour met computer scientist John Mollis, who became OnsiteIQ's co-founder and chief technology officer. Their platform automatically positioned walkthrough imagery on a digital floor plan. A remote viewer could move through the site, revisit the same place at an earlier date, annotate a condition and share it. The result resembled Street View for a building that had not yet finished becoming itself.
The local company with national customers
Early adoption required champions. New Line Structures gave the young company a chance on a 430,000-square-foot New York project and supplied feedback. OnsiteIQ entered the MetaProp accelerator looking for mentors, pilots and strategic capital. Khosrowpour said the program helped the company build a sales operation, multiply outbound activity by nine and sales by six. Mentor Joe Charczenko opened doors among insurance carriers.
That was useful, but it did not solve the map. OnsiteIQ could capture projects in New York. The owners and funds it hoped to serve built across the country. The startup could not win a broad enterprise contract while it remained local, and could not sensibly build a national operation without enterprise revenue. Khosrowpour's account of the solution is wonderfully impolite to cautious planning: they signed an enterprise customer and then hurried to become the company they had sold.
He has said the company bled as it scaled. Local field operations had to appear in new markets quickly while capture quality remained consistent. That qualifier matters. A patchy archive does not become better data by becoming larger. OnsiteIQ's advantage depended on making each walkthrough boringly dependable.
The company won the 2018 COINS Grand Challenge's New & Emerging Technology category, raised a $4.5 million Series A announced in 2020 and a $14 million Series B announced in 2023. In a more recent founder update, Khosrowpour said OnsiteIQ had operated in 270 cities and served more than 3,000 projects representing over $40 billion in construction.
Follow the consequence, not the hard hat
OnsiteIQ's customer focus also evolved. The most visible people on a jobsite are contractors, so construction software often begins with them. Khosrowpour became more interested in owners, developers and investors. A large project may be divided among dozens of trades and companies, each with its own incentives and narrow view. The economic consequence eventually gathers with the people financing and owning the asset.
“The industry is very fragmented, but at the investor level, it all becomes consolidated.”Ardalan Khosrowpour
That observation is both a sales strategy and a theory of information. The person paying for the building needs a coherent account of progress, but receives it from a system designed around fragments. A neutral visual record gives the owner another way to inspect the story. It can help check trade progress, spot early delay indicators and preserve a dated account when memories become conveniently flexible.
There is a social risk in this much visibility. Documentation can feel like surveillance with better typography. Khosrowpour encountered the resistance early: some management teams were wary of creating a record of every condition. His answer has been to frame transparency as shared reality. The useful version of accountability is not a theatrical hunt for blame. It is a room in which fewer people can profit from pretending the wall is finished.
The founder's reusable playbook
Find the layer where a fragmented market reconvenes. Standardize the raw evidence before automating its interpretation. Expand only if the data quality travels with you. Then give decision-makers a view that matches the scale of their responsibility.
Nine unglamorous years under one prompt
OnsiteIQ began with visual documentation, introduced progress monitoring in beta in 2021 and launched what it calls its Construction Intelligence platform in 2022. The sequence explains more than the labels. First build the record. Then compare it. Then analyze it. The artificial intelligence arrives after the reality has been made legible.
In February 2026, Khosrowpour announced an OnsiteIQ MCP, a connector that lets customers use their own AI assistants to question project and portfolio data. A team can request a report across many jobs, examine one asset, compare construction progress with a payment application or reconstruct how work unfolded floor by floor. A task that once demanded days of field visits and desk work can begin as a typed question.
The prompt is the smallest part of that achievement. Underneath it sit years of recurring captures, mapped locations, detections, progress tracking and aggregation. Khosrowpour called them nine unglamorous years. That phrase is refreshingly useful in an AI market attracted to instant revelation. A model can write an elegant answer in seconds. It cannot travel backward and create the missing Tuesday when nobody walked the fifth floor.
Visual Documentation launches.
Progress Monitoring enters beta.
Construction Intelligence platform launches.
OnsiteIQ MCP connects the project record to customers' AI assistants.
Khosrowpour now also advises early-stage deep-tech founders at Cornell Tech. It is a neat return to the institution where someone once let a late candidate recover his breath. He has spoken to students at Columbia about AI-assisted project management and the effects of abundant computing on project delivery. The argument remains grounded in the same question that bothered him in graduate school: why should an old industry continue making decisions with such poor visibility?
Construction is unusually honest about the distance between an idea and a finished thing. Renderings can float. Buildings have to stand up. Khosrowpour's work applies that same demand to construction data. The record should have a location. The claim should have a date. The picture should have something to compare against.
A project that can remember is still capable of delay, disagreement and surprise. It is simply harder to make those problems disappear into a camera roll. For an engineer who grew up around jobsites, that may be the point: give the building a memory before asking the machine for an opinion.