FCP Let’s talk

AI product leadership · Agent harness engineering

Felipe Chavarro Polanía

Build the system.
Earn the trust.

I connect AI process automation and product strategy with the engineering around the model: evaluation, control, and evidence that make complex systems easier to trust.

Explore my work CV & professional profile
Felipe Chavarro Polanía

Enterprise experience at HPE.
Independent research. Open-source tools.

THE ENGINEERING AROUND AI
Intent & constraints
AGENT HARNESS
ContextTools
Modelreason · propose · act
ControlEvaluation
Traceable outcomes
A working lens: define the task, constrain the action, inspect the evidence.
PRODUCT JUDGMENTSYSTEMS ENGINEERINGRESEARCH DISCIPLINE

01 / SELECTED WORK

From an idea to
inspectable evidence.

My focus is the layer between a promising model and a useful system: how it is evaluated, governed, and connected to real decisions.

OPEN SOURCE · PYTHON

Governed AI Systems

Research tools for deciding what to test, which experiments earn more compute, and whether LLM-judge comparisons preserve response identity.

Explore the repository
  1. 01

    Recipe discovery

    Fractional-factorial designs and descriptive contrasts for structured experiments.

  2. 02

    Compute governance

    Predeclared promotion policies and deterministic decisions over complete experiment cells.

  3. 03

    Judge audit

    Identity-aware evaluation that separates decision changes from display-label and parser failures.

Research and decision-support tools. Repository examples are synthetic; empirical findings live in the linked preprints.

AI PROCESS AUTOMATION

From workflow to prototype.

I created an AI-powered software factory that automates software development and enables Product Management to build robust prototypes in weeks, accelerating time-to-market and time-to-value.

ENGINEERING

Make the behavior inspectable.

My current focus is agent harness engineering: context, tool use, evaluation, and control around AI models, with clear boundaries on what the evidence supports.

02 / PUBLIC RESEARCH

Small experiments.
Better-grounded decisions.

Two sole-authored arXiv preprints on experimental design and compute allocation for micro-pretraining. Read the methods, artifacts, and limitations alongside the results.

These are public preprints. Their listing here does not imply peer-reviewed publication.

03 / PERSPECTIVE

Technical depth.
Product perspective.
Human responsibility.

I work where engineering decisions meet business needs and human consequences.

My background brings together enterprise technology at Hewlett Packard Enterprise, AI service development, and independent research. Today, I’m especially interested in how agent harnesses turn model capabilities into systems people can inspect and improve.

I care about responsible AI as an engineering practice: explicit assumptions, reproducible experiments, and honest limits on the claims we make.

View my professional profile

04 / GET IN TOUCH

Working on AI
that needs to work?

Let’s talk about AI products, enterprise solutions architecture, agent systems, and evidence-driven engineering.

Connect on LinkedIn Explore my GitHub Professional profile