Rilton Franzone was already debugging production systems used by real customers at the tender age of 17. There were no safety nets, just unclear requirements and live code with the expectation that things should work. That was where he learned to be an engineer.
Franzone began learning programming at 15, with online courses from Harvard, MIT, and HKUST. He managed to find a full-time role at a startup within two years as a working engineer. He quickly became responsible for shipping features and maintaining production systems.
His early experience influenced how he deals with problems to this day. He didn’t learn engineering in theory first and practice later. He learned by being useful, by building things that had to work, and improving them when they didn’t.
A career built on usefulness
Rather than specializing early, Franzone moved across whatever the product needed. Backend, frontend, mobile, data engineering, and AI. Each role was less about choosing a lane and more about solving the problem at hand.
Before joining Midpage.ai, he worked across fintech, academic research tooling, mobility, logistics, and SaaS, contributing to companies including WithClutch, CatalyzeX, The Drivers Cooperative, LetsLunch, and CodeGem.
At WithClutch, he helped build infrastructure capable of processing millions of loan applications, supporting hundreds of credit unions across the United States, systems where reliability wasn’t optional.
While employed at CatalyzeX, he developed large-scale web crawlers indexing over 400,000 code implementations from machine learning papers, serving a global research community of more than 200,000 users.
His technical stack reflects flexibility: TypeScript (Node, React, Next.js) and Python, which he uses as tools to build systems that are scalable, reliable, and aligned with user needs.
Joining midpage at the moment it mattered
In 2025, Franzone joined Midpage.ai as its third engineer, just before the company closed its seed round. It was a defining moment, one where engineering decisions directly shaped business outcomes.
In a three-person engineering team, there is little room for narrow roles. Franzone worked across infrastructure, frontend and backend systems, AI engineering, and evaluation frameworks; whatever was required to move the product forward.
He made a significant contribution to the development of Midpage’s legal AI research agent, which is now also the platform’s most widely used feature. The system, which is used by more than 300 law firms across the United States, helps lawyers deal with complex legal questions, retrieve relevant case law, and base their output on structured evidence. Attorneys can integrate it directly into their workflows.
Attorneys must be able to produce accurate output and defend it. Systems must be correct, but must also be able to explain why they are correct.
The quality of Midpage’s system has been externally validated. According to VLAIR’s benchmark from vals.ai, its legal research agent ranks among the top three legal AI systems globally, outperforming both ChatGPT and a lawyer baseline across multiple evaluation metrics.
Besides working on core product development, Franzone has also led integrations with major technology partners, such as Perplexity, Litera, Anthropic, and OpenAI.
That engineering work translated directly into company growth.
Measuring what matters in AI
More recently, Franzone has led the development of benchmark.midpage.ai, an evaluation framework designed to measure how well advanced AI systems handle complex, long-horizon legal tasks.
The system was built in collaboration with legal professionals and includes a custom dataset and a structured judging system that is capable of assessing accuracy and relevance.
This addresses the gap of evaluating AI models in real-world, high-stakes scenarios.
For Franzone, benchmarking is central to building reliable AI systems. You need strong evaluation to distinguish between outputs that are fluent and those that are genuinely correct. This is an important distinction in legal AI.
Working where the path isn’t clear
Franzone sees ambiguity as a difficult part of his work.
He learned early on to assess situations independently, take responsibility, and move forward even when the path wasn’t obvious.
This became the approach he repeated over time. He would identify where the system is constrained, apply technical leverage, and improve it incrementally.
Franzone’s objective always remained to build something that works, and improve on it, whether it’s designing APIs, debugging data pipelines, optimizing system performance, or transforming experimental AI workflows.
Building for high-stakes AI
Franzone focuses on AI systems used in high-stakes legal and other professional environments, especially in legal and document-heavy domains where correctness and trust are important.
Failure has real consequences in these contexts. Systems have to deliver accurate outputs, provide supporting evidence, and behave reliably under pressure.
Franzone’s approach reflects that reality. He stays close to users, ships production systems, learns from feedback cycles, and builds scalable products.
It is a philosophy shaped less by theory and more by experience, by years of working in environments where the work has immediate, measurable impact.
As the legal AI space continues to evolve, the demand for engineers who can bridge deep technical expertise with practical application is only increasing. Franzone’s career has consistently moved in that direction: toward building systems that are not only advanced but genuinely useful.
He started by learning through building. He’s still doing the same thing, just at a scale where the systems matter more, and so do the outcomes. Follow Rilton Franzone on LinkedIn or GitHub.
