Jeevy Fabrication Is Bringing AI to the Welding Shop Floor

The Macro: Industrial Fabrication Runs on Spreadsheets and Phone Calls

Custom metal fabrication is a $300 billion global industry that still operates like it is 1995. When a company needs complex weld work done, whether that is aerospace piping, industrial skid assemblies, or data center infrastructure, the process typically goes like this: an engineer sends out manufacturing drawings to a handful of shops they already know. Each shop manually reviews the drawings, estimates materials and labor, and sends back a bid. The engineer compares bids on a spreadsheet, picks a shop, and then spends weeks going back and forth on scheduling. The whole cycle takes anywhere from two to eight weeks for a single job.

The inefficiency is staggering when you consider the volume. Thousands of fabrication shops across the United States, most of them small to mid-size operations, are constantly trying to fill their capacity while engineers at OEMs and contractors are constantly trying to find available shops. It is a matching problem buried under layers of manual process, tribal knowledge, and fax machines that somehow still exist in 2026.

The construction tech and industrial software space has seen real investment in recent years. Companies like Procore and PlanGrid (acquired by Autodesk) have digitized parts of construction management. But the fabrication layer, the actual making of custom metal components, has been largely untouched by software. Part of the reason is that every job is different. This is not standardized manufacturing. A pressure vessel for an aerospace application has completely different specs, certifications, and tolerances than a structural steel frame for a data center. The matching logic is genuinely complex.

A few companies have tried marketplace approaches for manufacturing, like Xometry and Fictiv, but those tend to focus on CNC machining and sheet metal. Complex weld fabrication, the kind that requires certified welders, specific alloy expertise, and ASME code compliance, is a different animal. It has been waiting for someone who actually understands the work to build the software.

The Micro: SpaceX Meets AI Meets a Welding Shop in Cleburne, Texas

Jeevy Fabrication (Y Combinator S25) was founded by brothers Jeevesh and Vinay Konuru. The backgrounds here are relevant. Jeevesh was an engineer at SpaceX, where he led an eight-person team building cryogenic loading systems and managed a $110 million budget for Starship propellant systems design and procurement. He has an MIT BS/MBA and a Stanford MS. Vinay was previously at Rondo Energy as the personal engineer to the founder, and he won the Princeton Research Day 2023 prize for a direct air carbon capture system. He dropped out of Stanford’s Master of Engineering program to do this.

These are not software people who decided manufacturing looked interesting from a pitch deck. They are engineers who lived inside the procurement and fabrication process and saw how broken it was firsthand.

The product does two things. First, it acts as a fabrication contractor itself. The company operates out of Cleburne, Texas, and handles aerospace piping, skid assemblies, custom engineered fabrications, tooling design, field installation, steel manufacturing, and data center engineering work. They have a physical shop and they do physical work.

Second, and this is the AI play, their software converts manufacturing drawings into accurate bills of materials and resource-loaded schedules. That conversion step is where most of the bid cycle time lives. An experienced estimator might spend days reviewing a set of drawings, identifying every component, calculating material quantities, and building a schedule. Jeevy claims their AI reduces that bid and scheduling time by over 80%.

The combination of being both a fabrication shop and a software company is unusual. Most startups in this space pick one lane. Jeevy is using the shop as both a revenue source and a testing ground for the software. Every job they run through their own shop generates training data for the AI. That feedback loop is hard to replicate if you are just building software from the outside.

The website emphasizes quality assurance, safety protocols, and environmental responsibility. These are not buzzwords in fabrication. They are requirements. If your welds fail on an aerospace piping system, people can die. The certifications and compliance standards in this industry are demanding, and shops that cannot meet them do not survive.

At three people, the team is lean. The address in Cleburne puts them in the DFW industrial corridor, which is smart. Plenty of shops, plenty of demand, reasonable cost of operations.

The Verdict

I think Jeevy is building in one of the most underserved corners of industrial tech. The market is huge, the incumbents are slow, and the founders have the domain credibility to actually sell to fabrication shops and OEM procurement teams who are deeply skeptical of software companies that do not understand their work.

The dual model, running a shop while building AI software, is both the strength and the risk. It is the strength because it gives them proprietary data and credibility. It is the risk because running a fabrication shop is capital-intensive, operationally demanding, and a completely different management challenge than building software.

At 30 days, I would want to see how many external jobs they have run through the AI estimating tool and whether the 80% time reduction holds up across different job types.

At 60 days, the question is whether they can start licensing the software to other shops without cannibalizing their own contracting business. The platform play is where the scale lives.

At 90 days, I would want to see the first signs of a network effect. If enough shops are on the platform, engineers can post a job and get matched to the right shop automatically. That is when the marketplace dynamics start to compound.

The fabrication industry is not going to get disrupted by a pitch deck. It is going to get improved by people who know how to weld, know how to estimate, and know how to write software. Jeevy appears to check all three boxes.