Razvan Nistor

Physics Data & AI

Razvan Nistor

AI strategy, engineering, and teams.

I play guitar like Andrej Karpathy and do AI like Eric Clapton.

What I do Get in touch

About

Physics, Data & AI

Portrait of Razvan Nistor

I took my background in computational physics and applied it to Data & AI for the past decade. The most important thing I've learned is how to talk to the business and actually deliver meaningful value from technical projects.

Today, I lead an enterprise Data & AI practice. I work equally well with Engineers and Executives. My main focus is to keep it human, and to continue building up the right people and process to solve problems in this AI era and beyond.

I built this site to share my work and what's behind building AI platforms and teams that deliver. If that's your world too, I'd love to connect.

  • PhD, UWO 2009. Postdocs at IBM and Columbia 2009–2014.
  • Outside work: soccer, guitar, photography, family.

What I do

Connect strategy with delivery.

My focus for 2027: bring AI capability, forward deployed engineering, and trusted data foundations together around useful work.

  1. AI strategy & centers of excellence

    Set direction. Build capability.

    Connect business priorities with a practical AI roadmap. Build an AI Center of Excellence (CoE) around useful work: governance, coaching, champions, and maintained assets people can reuse.

    Make ownership, evaluation, and reviewable agent activity part of how applications are built and operated.

  2. Forward deployed engineering

    Stay close to the work.

    Work alongside users to understand the process before choosing the technology. Forward deployed engineering (FDE) connects discovery, hands-on engineering, and adoption around a real business problem.

    Simplify the workflow, combine code and AI where each helps, and deliver a useful change within the systems people already use.

  3. Data products & platforms

    Give useful systems a trusted foundation.

    Build the platforms, pipelines, and owned data products that support analytics, predictive models, and AI applications. Connect reliable data with shared business definitions, metrics, and relationships.

    Make quality, access, and lineage explicit, so people can understand an answer, trace its sources, and maintain the system behind it.

One connected practice

The CoE develops capability and shared standards. Work in the field tests them with users. Data products and platforms support reliable delivery. What we learn goes back into the practice.

My approach

Start with the work. Keep learning.

Stay close to users, make a useful change, and leave people with the skills and systems to carry it forward.

Start with the work

Understand the problem first.

Observe the workflow with the people who do it. Agree on what is worth changing, what good looks like, and which data and systems are available.

The useful outcome

A shared problem, an owner, and a way to measure progress.

  1. Map the work, handoffs, and exceptions.
  2. Establish a baseline and a useful measure of success.
  3. Agree on an owner, a bounded scope, and permitted access.

Ready to move forward whenThe people doing the work recognize the problem and the proposed outcome.

Make the idea tangible

Build a useful slice.

Simplify the process before automating it. Use rules, data engineering, models, AI, and human judgment where each earns its place. Test a small working solution with users.

The useful outcome

A working slice with evidence people can review.

  1. Build the data and integrations the workflow needs.
  2. Evaluate real tasks, exceptions, and access boundaries.
  3. Compare the new approach with the current way of working.

Ready to move forward whenUsers and reviewers agree that the change is useful and ready for a bounded rollout.

Make it work in practice

Build for the people who inherit it.

A useful application needs adoption and an operating owner. Make agent actions reviewable, keep decisions with the right people, and give maintainers the context to support the system.

The useful outcome

An adopted workflow, clear ownership, and a support path.

  1. Monitor quality, usage, cost, and failure modes.
  2. Exercise human review, fallback, and recovery paths.
  3. Coach users and rehearse the handover with maintainers.

Ready to move forward whenThe people using and maintaining the system can operate it with confidence.

Connect delivery back to strategy

Make the next build better.

Bring field experience back into the CoE. Improve the shared methods, maintain the useful assets, and test whether a data product or pattern can serve another workflow.

The useful outcome

A maintained asset and a better-informed next decision.

  1. Review what changed for users and what still needs work.
  2. Package useful patterns with examples, checks, and an owner.
  3. Prove reuse before expanding the scope.

Ready to move forward whenAnother person can use the pattern without depending on its original author.

Get in touch

Let’s keep it human.

On AI strategy, data platforms, or taking a useful idea into production. If that’s your world too, let’s compare notes.

Connect on LinkedIn