The corporate filing cabinet never vanished. It learned to sync. Contracts settled into SharePoint. Reports multiplied across drives. Email attachments became archives, then evidence, then mysteries. For most of enterprise software’s modern history, the structured database received the dashboards while the ordinary file received a hopeful name and another version number. Kyle DuPont built Ohalo in the neglected territory between those two worlds.
DuPont is an Atlanta-based founder with a career that has run through Tokyo, London, and institutional finance. He worked in finance and technology at MUFG and Morgan Stanley, and his own account of the period is charmingly free of cinematic revelation. More than fifteen years ago, while working in mergers and acquisitions, he read documents and pulled out details for reports. Market conditions, management quality, regulation, risk: the valuable bits were present, but a person had to find them.
This was the sort of labor that teaches an analyst two things. First, context matters. Second, context is expensive when it hides in prose. The lesson waited. DuPont moved through transaction-services work at ERM, served as COO at MatrixVision, and spent a year as a Fellow at the fintech investor Anthemis Group. He caught what he later called the startup bug. An early idea concerned anti-money-laundering workflows. The larger opportunity emerged from the documents beneath the workflow.
The regulation was a door, not the room
DuPont co-founded Ohalo in 2017 with Alistair Jones. GDPR was approaching, and companies faced a brutally practical question: where was the personal information they were now expected to manage? A database could be queried. A warehouse could be cataloged. A directory containing years of PDFs, scans, presentations, and correspondence was another creature entirely.
Ohalo began as a GDPR compliance solution. Then its founders followed the problem beyond the regulation. Compliance had merged with security, DuPont explained in a 2025 interview, and the central issue was not merely a legal requirement. Companies did not understand their own data.
“The compliance issue merged with security. We realized the core issue wasn’t just regulatory, it was that companies didn’t understand their data.”Kyle DuPont
That distinction gave Ohalo room to grow. Its Data X-Ray platform was designed to discover, classify, extract, and redact information across unstructured sources. The name is unusually honest marketing. The product is supposed to show an organization what is inside the opaque body of its files, without asking a procession of humans to open each one.
DuPont likes an example with a human scale. A résumé is not merely a collection of names, dates, and email addresses. It is a career history. Traditional pattern matching can locate an address; it struggles to recognize what the document means as a whole. Ohalo’s wager is that machine learning, natural-language processing, optical character recognition, and large language models can supply that missing context at enterprise scale.
A sandbox with 600,000 chances to be wrong
The useful test arrived in 2019. Ohalo won the first Safetytech Accelerator Challenge, a competition that connected data companies with difficult industrial problems. The pilot involved a vast collection of incident reports held by Britain’s Health and Safety Executive. The reports needed anonymization before outside researchers could use the dataset. Historically, that meant slow, careful redaction by people.
Data X-Ray processed 600,000 records in 1.4 days, according to a later account from Lloyd’s Register Foundation, with reported accuracy of 99 percent. The manual comparison was 12.5 years. More important than the handsome arithmetic was the feedback loop. Ohalo installed a server inside the customer’s environment. Data scientists reviewed missed material and false positives, then supplied examples that improved the models.
The pilot became a product workshop with consequences. DuPont said it opened relationships with construction companies Costain and Wood Group and helped Ohalo extend Data X-Ray’s capabilities. A controlled challenge had done the rarest thing in enterprise innovation: it let a startup learn on real material without pretending the laboratory was the market.
It also revealed DuPont’s preferred style of ambition. He talks about policy, scale, and AI, but the claims return to work that can be counted. Files scanned. Words processed. Sensitive passages identified. A migration completed before a deadline. The abstraction must eventually survive contact with somebody’s untidy folder.
The analyst’s revenge
Years after those M&A assignments, DuPont appeared in an Ohalo demonstration with a familiar kind of document. He pointed to material about market conditions, management quality, regulatory environment, and risk factors. His younger analyst self would have extracted it by hand. Data X-Ray’s Extractors could make it queryable.
The moment closes a satisfying loop. Founder stories often give the first job a ceremonial role: a hardship survived, a credential acquired, a boss escaped. Here the old work remained technically useful. DuPont knew what a diligent reader wanted from a document because he had once been that reader. Product intuition arrived disguised as clerical fatigue.
“A CV isn’t just a name and an email. It’s a career history. You can’t identify that meaningfully with pattern matching alone.”Kyle DuPont
The company has since applied the same principle to larger estates. In one bank divestiture described by Ohalo, the task involved tens of millions of files and a legal deadline. Data X-Ray ultimately scanned more than 100 million files to help determine what information belonged on each side of the transaction. DuPont framed manual review at that scale as an intractable human problem, because the files keep changing even while the review is underway.
This is enterprise software at its least photogenic and most legible. A deal cannot wait years for a file review. A regulator does not accept “the shared drive was complicated” as a control. The value is not a delightful click. It is the absence of a terrible month.
AI moved the perimeter
Generative AI made Ohalo’s original problem newly fashionable. Companies now want models and agents to retrieve internal information, summarize contracts, answer operational questions, and automate decisions. Yet the model cannot responsibly use what the organization has not classified. A forgotten archive may contain the best answer and the one document the user should never see.
DuPont argues that the old security perimeter has dissolved. Cloud storage and distributed work weakened the neat boundary around a company network. In his formulation, the perimeter is now identity and the data that identity can access. A job title alone is too blunt a permission system. The sensitivity, intellectual property, trade secrets, and regulatory meaning of a document should help determine access.
“The perimeter is no longer the edge of a firewall. It’s the identity and the data that identity has access to.”Kyle DuPont
It is a sternly practical view of AI readiness. The glamorous layer is the model. The durable layer is the inventory: what exists, where it lives, what it means, who owns it, and who may use it. DuPont’s public commentary keeps returning to files and folders outside the systems most data teams watch. He is less interested in giving an agent another trick than in making sure it understands the room it has entered.
Work in Tokyo and London, including MUFG and Morgan Stanley, builds familiarity with institutional systems and document-heavy analysis.
An Anthemis fellowship helps DuPont develop the business that becomes Ohalo.
Ohalo is founded around privacy compliance and the larger problem of unstructured data.
A Safetytech pilot processes 600,000 records and creates a live feedback loop for Data X-Ray.
DuPont’s focus broadens toward governed data as the foundation for enterprise and agentic AI.
A trans-Atlantic company for an everywhere problem
Ohalo grew between London and Atlanta, with DuPont publicly based in the Atlanta metropolitan area. The company works with banks, governments, manufacturers, and other organizations whose data estates are both enormous and difficult to move. Its partnerships have stretched from Collibra to Macnica, and a 2025 collaboration in Japan connected Data X-Ray classification with Microsoft Purview workflows.
DuPont’s own geography fits the company. His public profile lists English and Japanese at native or bilingual proficiency, with French at limited working proficiency. He has moved among different financial and technical cultures long enough to recognize a universal corporate dialect: the document whose author has left, whose owner is uncertain, and whose contents remain important.
Ohalo says its mission is to create order from data chaos. The phrasing is tidy; the work is not. Classification policies collide with local practice. Old systems linger. A scanned form resists extraction. A false positive irritates a reviewer. An AI initiative creates a new appetite for material nobody planned to expose. DuPont’s career suggests that the mess is not an embarrassing prelude to the real work. The mess is the work.
DuPont discusses Ohalo’s origin, unstructured data, generative AI, and security on Cyber Security America.
Play the 40-minute interview ↗There is a quiet wit to building a sophisticated AI company around the digital equivalent of a cabinet nobody wants to open. Yet the timing is serious. Agents need context. Security teams need evidence. Data leaders need to know whether a useful file is safe, current, and allowed. The better the model becomes, the less charming it is to discover that the underlying information was never governed.
DuPont began with the analyst’s irritation of reading everything himself. He arrived at the founder’s conviction that organizations should understand the information they already possess. Between those points sit regulations, pilots, millions of files, and a company that kept widening its answer without abandoning the original question. What is in the document? It sounds small. At enterprise scale, it can occupy a career.