A small team emerging from Y Combinator has built software that, according to its founders, promises to compress weeks of engineering document review into hours, catching the kinds of design inconsistencies that can derail multimillion-dollar construction projects before ground is broken.
Torus, a five-person company from a recent YC batch, says it is working with Fortune 500 firms on an AI agent that cross-checks piping diagrams, spec sheets, and heat balances for data centers and energy infrastructure. The platform flags mismatches—a valve rating that contradicts a design parameter, say, or electrical loads that don't square with transformer capacity—and drafts the follow-up paperwork: requests for information, procurement approvals, transmittals.
The thesis is straightforward. Engineering firms and project owners spend thousands of hours reconciling specifications across capital projects, hunting for errors that would otherwise surface as expensive change orders once contractors are on site. Torus aims to automate that reconciliation work, or at least a meaningful slice of it.
Running Checks Across Fifty Document Types
According to the company's website, Torus runs consistency checks across more than fifty categories of project documents. The system ingests piping and instrumentation diagrams (P&IDs), material balances, and datasheets, then scans for discrepancies. If a design pressure rating doesn't match a relief valve setpoint, the software surfaces it.
The platform goes further than error detection. It generates draft emails, routes approvals through internal chains with an audit trail, and integrates with collaboration tools like Microsoft Teams. The system tracks decisions and action items across threads involving multiple parties, which in theory keeps everyone aligned as a project evolves.
"We can help you do weeks of work in hours," co-founder and CEO Marcus Lima wrote on LinkedIn. "Torus can understand a project, including complex diagrams, and generate full engineering packages."
Founders With Prior Ventures
Lima previously started Heimdal, a direct air carbon capture company, and studied at Oxford and Cambridge, according to his YC profile. Co-founder and CTO Rahul Thayil built Orion, an AI threat intelligence platform that raised a seed round.
The startup is now hiring electrical engineers in San Francisco and Houston, software engineers in San Francisco, and physical infrastructure engineers in San Francisco. Salaries range from $100,000 to $200,000, per the Y Combinator directory.
Two Deployment Paths
Torus offers clients a choice. The managed version runs in the vendor's cloud with data isolation; the company says it deletes customer files after each request and never trains its models on proprietary information. The sovereign version runs entirely inside a customer's virtual private cloud or on on-premises servers, so project data never touches the public internet.
SOC 2 Type 1 and Type 2 certifications are in progress, the company said, though neither has been completed.
Torus sells through pilot programs. Prospective customers submit five or six project documents for a 15-minute demonstration of what the system catches.
Crowded Field, Growing Appetite
Torus is wading into a construction-tech landscape where AI adoption has accelerated. A recent Unanet survey of architecture, engineering, and construction firms found that a significant majority now use AI in some capacity, up sharply from the prior year. Yet confidence in underlying project data remains low, with fewer than three in ten respondents reporting high trust in their information.
The YC cohort that includes Torus features several other startups targeting engineering and design workflows. Structured AI is building what it calls an agentic workforce for construction design, focused on drawing quality control and learning firm-specific standards. Saphira AI tackles requirements management for safety-critical hardware, positioning itself as an alternative to legacy platforms like Jama and IBM DOORS.
Outside the accelerator, ALICE Technologies announced a partnership with McKinsey earlier this year to bring generative scheduling to capital projects. nPlan offers AI-powered schedule risk forecasting.
Y Combinator's Demo Day, typically a venue for customer wins and fundraising updates, is scheduled for September. Torus has not disclosed pricing or named the Fortune 500 customers it claims to be working with. The company's willingness to run pilot demonstrations suggests it is still refining its go-to-market motion, perhaps feeling out which segments of the infrastructure market will pay for document automation at scale.
Whether Torus can convert pilots into long-term contracts will likely depend on how well its consistency checks hold up against the Byzantine complexity of real-world capital projects—and whether engineering firms trust an algorithm to catch what human reviewers might miss.
