Ai Leadership & Education

Contracts Connected builds software for construction companies — the contracts, certificates of insurance, and compliance paperwork that follow a project from groundbreaking to closeout. Their customers live in documents where a missed detail isn't an inconvenience; it's a liability.

They came to Nullwest with a question a lot of software companies are asking right now: how do we actually implement AI? Not a demo, not a pilot that dies in a slide deck — AI working inside a real product, handled by their own team. That meant the engagement had to produce more than features. It had to produce a team that could keep building after we left.

Contracts Connected chose Nullwest for its AI implementation experience — and for an unusual proposal: instead of building AI features off to the side, we would train their existing engineers to build them.

Dave, who holds a PhD in Digital Media from Georgia Tech and has taught at Georgia Tech, The New School, CUNY Staten Island, and Cornish College of the Arts, led the education program. Scott, a former Apple AI engineer, led the build — and taught alongside Dave, fielding the hard AI questions as they came up in their team's real work. Over three months, Nullwest:

- Mapped the entire product base and roadmap with their leadership, prioritizing AI needs against real production timelines - Ran one-on-ones and group classes covering Python fundamentals, OCR, RAG search, AutoGen, and agentic AI systems - Reviewed and updated their codebase, where Scott stood up new AWS infrastructure managed with Terraform - Built a working agentic RAG pipeline in Pinecone — Scott's build, with chunking strategies tuned to construction insurance documents like certificates of coverage and certified payroll — as a foundation their team could extend

The work was deliberately scoped: build enough to show how it works, then hand the keys to their engineers. That's not a compromise — it's the capability-over-dependency principle Nullwest was founded on.

Most AI consultants build you a feature. Then they leave — and so does the knowledge. The expensive failure mode isn't bad AI; it's AI your team can't maintain, extend, or debug. We treated education as the deliverable. Dave brought a decade-plus of university teaching to structured one-on-ones and group classes, while Scott answered the practitioner questions — the ones that only come up when real code meets real documents — taking their engineers from Python fundamentals through OCR, RAG, and agentic systems fast, because the curriculum was built around their actual product instead of toy examples.

You can't bolt AI onto a codebase that isn't ready for it. AI features fail quietly when the foundation underneath them is improvised. Before any model touched production, we reviewed and updated their codebase, optimized the slow parts, and Scott rebuilt their infrastructure on AWS with Terraform — so every AI workflow their team builds lands on ground that's reproducible and theirs.

A demo isn't a foundation. It's easy to make a RAG pipeline look good on stage; it's harder to make one a team can build on. Scott built a working agentic RAG pipeline in Pinecone with chunking strategies designed for the documents construction companies actually handle — certificates of coverage, certified payroll — then stopped, on purpose, while their engineers took it forward. The point was a starting line, not a dependency.

In three months and roughly 100 hours, Contracts Connected's engineering team went from asking how to implement AI to shipping their own AI-enabled workflows: ingesting uploaded certificates of insurance, running OCR, structuring the extracted data, and answering questions against it through chat. By our estimate, hiring an outside team to build all of it directly would have taken five to ten times the effort — and left them dependent on that team forever. As we put it: we trained them to replace us.

[Optional: client quote here]

Trying to add AI to your product — and wondering whether your own team could build it? Nullwest's founders have shipped AI in production from Apple to Xembly, and taught it at four universities. Let's talk about yours.