Kontratar AI
AI-native B2B Contract Tooling
Designing an intelligence platform that makes government contracting faster, compliant, and built for how teams actually work.
AI / GovTech / B2B
Industry
iOS, Android, Web
Platform
Project Overview
Government contracting has always been a game of precision. Miss a requirement, lose the bid. Spend weeks building a proposal, and it still might not win. The teams doing this work are sharp, but the tools and processes around them weren't keeping up. I worked on Kontratar AI as a significant contributor alongside a cross-functional team, focused on translating complex AI capabilities into a product experience people could actually trust and use.
The Challenge
The proposal process for government contracts is slow by design, and that's before you account for how most teams actually work. Opportunities are discovered late. Requirements get misread. Teams collaborate across disconnected documents. And a single compliance gap can disqualify months of work. The business cost is real: lost contracts, inefficient teams, and resources spent on bids that never had a chance. But the deeper problem isn't just time, it's trust. Teams weren't going to hand their proposals to an AI system unless they were confident it understood compliance as well as they did.
"In government contracting, speed alone isn't the value. Speed with guaranteed compliance is."
My Role & Approach
What I owned
I came into this project focused on one core question: how do you make AI feel reliable in a domain where the stakes are high and errors are disqualifying? My work covered the end-to-end experience, opportunity discovery, proposal generation, team collaboration, and compliance review.
How I approached it
I treated this less like a feature design problem and more like a workflow design problem. The AI capability was already there. The design challenge was embedding it into a process that felt structured, transparent, and human enough to actually be used.
The core of the product — where AI reads solicitation documents and produces structured, compliant proposal drafts ready for team review.
Traceability between requirements and outputs
Every section of a generated proposal is linked back to the specific requirement it addresses. This wasn't just a UI feature, it was a trust mechanism. Users needed to see that the AI had actually read and understood the solicitation document, not just produced plausible-sounding text.
One unified system across the full lifecycle
Discovery, writing, collaboration, compliance review, and tracking all live in the same platform. Unifying the workflow introduced some initial complexity in system design, but it eliminated the handoff failures and version conflicts that were costing teams time and accuracy.
Compliance validation before speed
When we had to choose between faster outputs and more accurate compliance checking, we chose accuracy every time. A non-compliant proposal has zero value regardless of how quickly it was produced. The additional processing this required was a real trade-off, but it was the right one for the domain.
Surfacing AI guidance, not just AI output
Alongside the proposal generation engine, we built an AI compliance assistant that could answer regulatory questions, flag gaps in real time, and explain its reasoning. The goal was to close the gap between what the AI produced and what an expert would have reviewed.
Strategic planning before the bid, not after
Most proposal tools assume you've already decided to bid. The team simulation feature lets organisations model different team configurations, identify capability gaps, and assess their realistic chances before committing resources. For business development teams juggling multiple active opportunities, that upstream clarity is the difference between pursuing the right bids and spreading thin across the wrong ones.
Outcomes & Reflection
Kontratar AI is live and in use. But the metric I think about most isn't time saved, it's the shift in how teams approach the process. Proposal development used to feel like a high-stakes guessing game. The goal was to make it feel like a structured, repeatable system where the AI handles the heavy lifting and the team focuses on strategy and judgment.
This project reinforced that AI products live or die on trust. You can have the most capable model in the room, but if users can't see how it's thinking or verify what it's doing, they won't rely on it when it counts. The design work that mattered most here wasn't the interface, it was the logic, the transparency, and the structure that made the intelligence feel usable.
