By Akam Pachachi, founder and CEO of Clash Nexus AI
I didn't plan to start a construction software company. I've founded two tech companies before and sold them. After the second exit I was bored. Then I had a long conversation with a friend of many years who was a VP at a large construction firm at the time.
I asked him what actually costs his industry money that it shouldn't have to spend. It was the issues and clashes that get missed during preconstruction drawing review. The duct drawn through a beam. The detail that references a section that doesn't match. The piece of work that every drawing hands to "others" and no contract picks up. Each one is cheap to fix on paper and expensive to fix once crews and materials are on site.
One friend's opinion isn't a market, so I talked to more than 30 industry executives. About 80% of them told me the same thing my friend did. That was enough to start. A CTO I'd built with on earlier companies came on to lead development. We spent about eight months building before we launched on August 10, 2026.
The problem, in one paragraph
Construction teams still tender, price, buy out and build from issued PDF drawing sets. Those sets are reviewed manually, page by page and discipline by discipline, but the costly problems live between pages. The duct is drawn on the mechanical sheet and the beam on the structural sheet. Each is correct on its own, and nobody overlays the two. A schedule row is missing a size, and nobody catches it because the plan that references it looks fine. A 200-sheet set has just under 20,000 possible sheet-to-sheet pairs. No human checks all of them, and no one should have to. On top of that, drawings are full of notes like "by others," "delegated design," "NIC" and "connection by fabricator." Each of those is real work, and nobody prices it until it turns into a change order. When a revision arrives, the new sheets are usually compared with the old ones by eye.
How Clash Nexus works (high level)
You upload a complete PDF drawing package: architectural, structural, MEP, other trades, even shop drawings. No BIM model is required. That matters because many teams, especially trade contractors and precon groups at bid time, only have the PDFs.
From there, the platform does roughly what an experienced coordinator does, just systematically:
- Structures the package. It identifies sheets, disciplines, floors, buildings and scope, and works out how drawings relate to each other.
- Pairs the right sheets. Instead of comparing every page with every other page, it pairs sheets that should agree across disciplines, by floor, building and scope. Then it checks each pair for conflicts. The geometric cross-check is what catches things like HVAC against structure, plumbing against structure, or mechanical against architectural.
- Runs several review layers in one scan. Clashes, sheet-to-sheet contradictions, missing information (a schedule row with no size, a piece of equipment with no support detail), drawing-versus-spec conflicts, constructability and code signals.
- Shows its work. Every finding is pinned to the exact sheet and location, with the drawing evidence, a severity, the likely impact and a recommended next step. A reviewer can accept or reject each one. Reviewed findings can be sent as RFIs to Procore and Autodesk.
- Scope Ledger. It reads the hand-off notes off the drawings and builds one register of every handed-off obligation: what the work is, which sheet it's on, who it's handed to, whether the drawings give that party what they need, and whether a Procore bid package covers it (read-only).
- Revisions. When a new issue arrives, it compares it to the previous set. It shows what changed, which findings were resolved or are new, and which RFIs and scope items are affected.
Under the hood, drawings go through AI, OCR and document-processing services under business/API terms. Customer drawings aren't authorized for training general-purpose models. Each customer's projects are isolated to their own workspace, and everything is encrypted in transit and at rest.
What I've learned so far
1. "Issues" and "clashes" are different products. A lot of AI drawing review looks at one sheet at a time. That's useful, but the expensive problems show up when drawings meet. One prospect told us they'd used another AI review tool for about four years. It found issues but not clashes, and it was being shut down after an acquisition. The cross-discipline check turned out to be the thing people actually came for.
2. A drawing set is a graph, not a pile of pages. The naive approach compares everything with everything. For a 200-sheet set that's just under 20,000 comparisons, and most of them produce noise. The real work is figuring out which relationships matter before you check anything.
3. More findings isn't better. It's tempting to lead with a big issue count. Contractors have lived through BIM clash reports with thousands of irrelevant clashes, and they don't want another pile to wade through. What they need is precision, deduplication, sensible severity and evidence they can verify in seconds. We designed around one principle: evidence before promotion. A finding only surfaces when the drawings support it.
4. It's not "BIM clash detection without BIM." Model-based clash detection is excellent where coordinated models exist. Drawing review covers a different layer: the issued documents that people price and build from. The two work best together.
5. Some of the most expensive problems aren't drawn at all. Talking with estimators, I kept hearing about work nobody priced, like "who's designing the parapet clip anchors?" That became Scope Ledger. As I put it when we announced it: coordination catches what's drawn wrong, and Scope Ledger catches what isn't drawn at all.
6. Drawings are someone's IP. Construction teams are rightly careful with project documents. We built workspace isolation, encryption and a clear no-training policy from the start, not as an afterthought. We're working toward SOC 2 readiness but aren't certified yet.
For a sense of scale, here's one run from our own testing. On a 256-page drawing package, Clash Nexus flagged 65 design issues and 5 cross-page clashes for the team to review. That's one project, reported by us, not an industry benchmark. Your results will depend on your drawings.
Where we are and what's next
We're live. The team is small and based in Ontario, Canada, and we serve construction teams across North America. We built it for GC preconstruction, VDC/BIM, estimating and coordination teams, MEP and structural engineers, and trade contractors. Procore and Autodesk integrations are live, so reviewed findings can go straight into the RFI workflow a team already uses.
Try it on your own drawings
If you work in preconstruction, VDC, estimating or coordination, the most useful thing you can do is run a real project through it and tell me where it's wrong. Your first full scan is free and doesn't need a credit card: clashnex.ai. Paid plans start at CA$499/month. I read every piece of feedback: hello@clashnex.ai.
Vested interest disclosure: I, Akam Pachachi, am the founder and CEO of Clash Nexus AI (Thelon Technologies Inc.), the product described in this story. The one test result mentioned comes from the company's own testing.
Top comments (0)