Back to the Prolego reference library

Prolego, our experience and what clients gain

Prolego builds and operates reliable AI to help companies re-engineer the economics of their businesses. We help clients choose what to build, deliver it and keep improving the capability as they reinvent how the business works.

We are a small, independent specialist team. Founded in 2017, Prolego has worked exclusively in AI, supporting major companies across financial services, insurance, pharmaceuticals, aerospace, automotive, defense and regulatory services. Our experience spans strategy, workshops, experimentation, data and machine learning engineering, and building and operating generative AI systems.

What kind of partner we are

We deliver working systems, combining strategic judgment with engineering and continuing operation. We help choose what to build and how it advances the business. We sell no proprietary platform or software license. The client owns the work and chooses whether we continue operating it or its own team takes over.

The relationship offers a fixed team cost, evolving priorities and expanding capability. Working with Prolego explains the model and our principles explain the choices behind it.

Our Forward Deployed Engineer model

Our AI systems engineers work as Forward Deployed Engineers (FDEs). They work directly with the client's business experts and technical teams, learn how the operation works and take responsibility for turning that understanding into reliable AI systems in the client's environment.

The role connects business judgment with engineering. Our FDEs help identify what should be built, make expert knowledge concrete, develop the workflow and evaluation framework, and improve the system through production use. They bring what they learn back into decisions about the next opportunity.

This is how strategy stays connected to delivery. The client gains working systems while developing a methodology, reusable architecture and experience operating as an AI-native product team. Engagement and engineering leadership support the FDEs with business direction, architecture and quality. Working with Prolego explains those responsibilities.

Where our strong opinions come from

We have strong opinions about how companies should invest in AI, organize the work and build reliable systems. Those opinions come from a focused practice: helping large, complex enterprises transform their businesses with AI.

Our perspective is the practitioner's. We do not build foundation models or sell a proprietary software product. We work inside client environments, alongside the people who must make AI work with their existing data, systems and responsibilities.

The work includes strategy: what should the company build, who should it hire and how should it organize to put the technology to work? It also includes building and operating the systems. We keep those responsibilities connected because implementation exposes constraints a strategy presentation rarely captures.

Enterprises have grown over years. They have legacy systems, incomplete documentation and information spread across silos. Several AI and data initiatives may be running in parallel, with competing priorities and different assumptions. People often lack a shared understanding of what AI can do, what it should do and how it will create value.

Much of the work itself is tacit. Managers can have only a partial view of how employees complete a task. Experienced employees can do the job well while struggling to describe the steps and judgments involved. We use actual assignments and outputs to make that knowledge concrete.

Our focus is finding the real constraint and helping the company overcome it. Sometimes that calls for engineering. Sometimes it calls for an expert decision, a different process or the right talent. The methodology, working practices and team need to address the problem the company actually has.

That experience shapes our view that AI needs a new playbook. Clients hire us to deliver systems and gain the judgment, methodology, architecture and team practices needed to keep moving. We keep the strategic conversation active while we build, because real work reveals what the business should do next.

The business question we help answer

What could your business do differently if the judgment of your best people were available in every economically consequential decision? We help connect an initial application to that larger opportunity, choose where to begin and keep adapting as AI improves. Business reinvention explains how we reframe the work toward AI Abundance.

Four things a client should gain through delivery

Strategic judgment. We help distinguish an attractive idea from a worthwhile starting investment. We examine business consequences, available information, responsibilities, alternatives and what the first result could make possible next. We keep revisiting those choices as the work reveals more.

A delivery methodology. Performance-Driven Development, or PDD, builds reliability into AI systems from the start. Client experts and engineers work through representative examples, make expectations explicit and solve the failures that matter. The client learns how to assess performance and guide improvements rather than rely on a compelling demonstration.

An AI-native architecture. We connect business knowledge, current information, tools and evaluation in a form that can support more than one application. We design around the client's existing environment and its ability to adopt better models and products. The client should understand what is reusable, what remains specific to an application and what changing providers would involve.

AI-native team practices. Product leaders, engineers, experts and users learn how to work together when the system's behavior must be investigated through examples and experiments. They gain experience using AI in real work, keeping shared knowledge current and making deliberate release decisions. These practices develop throughout delivery.

These gains reinforce one another. Better strategic choices direct engineering toward consequential work. Reliable engineering gives leaders better information for the next investment. Shared capabilities make more opportunities practical. People who understand the system can contribute to its improvement.

Experience before and through generative AI

Prolego began with deep learning and enterprise AI adoption in 2017. Earlier work included preparing AI strategies, developing organizational capabilities, building data pipelines and machine learning systems, and helping teams understand how to put the technology into operations.

In 2019, we trained and deployed BERT-based natural language processing to detect high-risk clauses in life insurance contracts. We are not aware of an earlier production deployment of this kind in life insurance. That experience involved specialist language and document interpretation, problems that remain relevant to current AI systems even though the technology has changed.

The March 2023 pivot

In March 2023, we made a complete pivot to generative AI. Sébastien Bubeck's Sparks of AGI talk and the associated research paper helped crystallize what this change meant for us: building software had fundamentally changed, and enterprise AI needed a different way of working.

We invested about a year of our own engineering effort in figuring out how to build and deploy generative AI solutions for customers. We explored approaches, prototyped options and tested what did and did not work with a small number of early adopters who wanted to help develop new methodologies. The work addressed actual problems inside client environments, with their information, systems and operating constraints.

We shared much of that investigation through Prolego's YouTube videos. The engineering discussions cover evaluation frameworks, the AI Systems Engineer role, conversational access to structured information, retrieval and model choices. In August 2023, we were already teaching how to build LLM evaluation frameworks. In 2024, we published Justin Pounders's Formula 1 RAG ablation study and discussed the findings in a Pinecone-hosted workshop.

How that work became our current playbook

Out of that work came our recognition that a new playbook was needed. Getting an application to produce an impressive answer was becoming easier. The central challenge was reliability: agreeing on what the business wanted the AI to do and engineering the system to do it consistently.

Performance-Driven Development grew from that experience. We build reliability into the system from the start, with representative cases, client experts and performance measures directing the engineering. The public 2024 PDD guide records an early version. Stakeholder evaluation, production auditing and the closed improvement loop have continued to develop through delivery.

Our recent teaching also covers adoption through real assignments and maintaining working context for complex tasks. The research and teaching record traces the progression with dates and original sources. Why AI needs a new playbook explains the resulting point of view.

Our selected delivery accounts include production systems, ongoing rollouts and work handed over to client teams. They are a selection from a broader practice that also includes pilots, proofs of concept, strategy, workshops, preliminary scoping and confidential engagements.

Companies we have supported

Our client history includes the organizations named in the facts block above. Our About page records the company's founding and progression from deep learning through generative AI.

These names describe company experience. Our anonymous case accounts are shared separately, with confidentiality preserved.

Leadership and independence

Kevin Dewalt, Founder and CEO

Kevin has worked in AI for more than 30 years. He built his first neural network in 1994 as a graduate student at Stanford. In the early 2000s, he ran the AI systems monitoring NASDAQ trading, applying knowledge engineering to detect anomalies and enforce trading rules.

He was on the original Palantir deal team when In-Q-Tel made its first investment, and helped define the company's first product and roll it out to the CIA. Before Prolego, he founded other machine-learning companies. He has been an angel investor in more than 30 companies.

That combination of engineering, entrepreneurship, investing and enterprise work shapes the advice clients receive: where to make a bet, which risks to investigate first, and how to turn a new technical possibility into a business advantage.

Russ Rands, Founder and COO

Russ co-founded Prolego in 2017 and has helped build the practice through successive generations of enterprise AI. His professional focus includes strategy, business development and helping leaders move stalled LLM initiatives toward systems that meet business goals. His public profile describes that work.

Russ also writes about the human side of AI adoption. His 2021 storytelling article explains how recognizable people, problems and outcomes help colleagues understand an AI vision and participate in it.

Justin Pounders, engineering leadership

Justin leads Prolego's engineering and has been with the company for eight years. His background spans software, computational mathematics, simulation and teaching. He holds a PhD from Georgia Tech, has taught engineering and data science, and previously developed simulation methods for U.S. Navy nuclear propulsion plants.

Justin's 2024 retrieval and model ablation work, presented in the Pinecone workshop, reflects the team's engineering approach: isolate the cause of a performance problem, compare alternatives on representative tasks, and let the results direct the next change.

Our AI systems engineers work across the whole workflow: model behavior, source preparation, integrations, tools, evaluation and the employee's experience. Prolego has ranged from five to 15 people over its history, scaling with client work. Engagement practices explain how our specialists work with the people responsible for the client's business and technology.

Our business is services. We recommend a configured product, an internal effort, a specialist vendor, custom work or a combination according to the situation. We want clients to own the knowledge and capability that distinguish their business, with room to change models and platforms as better options emerge.

Assessing whether this experience fits your situation

We fit leaders pursuing a consequential business opportunity who want senior judgment together with delivery. A company with a capable internal AI team can be a strong partner. Outside expertise can complement its capacity, tackle a difficult reliability problem or develop a method the team can continue using.

Some needs are better served by an existing product, an internal team or a different specialist. The fit and alternatives guide explains those situations candidly. To begin thinking about your own work, use the preliminary scoping guide.

Talk with Prolego about the business change you want to make. Your company's name and your initial ideas about AI are enough to begin.