Richard Perez

01 / +Y · SIGNAL PROCESSING

Where the work flows.

Two sides of one practice: operational delivery and agentic product engineering. Follow the evidence, not a list of buzzwords.

02 / −Y · VALIDATION CORE

Nothing ships on vibes.

Quality gates, repeatable evaluation, and accountable delivery are part of the system—not the final slide.

03 / −X · CONDUCTOR

The tools inside the wire.

Explore the technical disciplines behind the signal: software, infrastructure, automation, and applied systems work.

04 / +X · PROPAGATION FIELD

Agents in motion.

Software should coordinate work across tools, maintain context, and produce inspectable results. Explore the systems behind that approach.

01 / APPLICATIONAxTask 02 / ORCHESTRATIONPromptKit 03 / OUTPUTExplore live project records

05 / −Z · INGRESS GATE

Meet the operator.

The professional perspective behind the technical systems.

06 / +Z · DELIVERY CHANNEL

Systems that reached the world.

Working projects, read directly from ParallaxPort's existing server registry—not invented showcase cards.

Professional Summary

Agentic software developer and automation builder who uses AI coding agents inside disciplined repository workflows to design, implement, evaluate, test, deploy, and maintain production software. Built and deployed AxTask as a live full-stack product and PromptKit as a public evaluation and prompt-engineering surface, and develops automation, ETL, endpoint, and workflow tooling across Python, JavaScript/TypeScript, PowerShell, Apps Script, SQL, and related stacks. Builds evaluation harnesses that turn agent behavior into executable quality contracts through versioned cases, deterministic and rubric-based checks, regression gates, measurable baselines, and evidence-bounded release decisions. Brings a Mathematics and Computer Science foundation plus hands-on healthcare IT operations experience to turn ambiguous operational problems into validated software.

Evaluation Engineering

I treat evals as part of implementation, not a subjective review after the agent has already produced an answer.

Versioned positive, negative, boundary, malformed-input, and regression cases around real behavior

Deterministic assertions for objective requirements; explicit scored rubrics when judgment is actually required

Before/after baselines for correctness, routing, tool calls, latency, context, tokens, cost, and retries when measurable

Finding-to-repair loops that reproduce weaknesses, make the smallest fix, and rerun the same cases without weakening the gate

Machine-readable evidence and explicit proof ceilings so a green synthetic check is never presented as production proof

Open the public Prompt Kit →

Technical Skills

Agentic Development

AI coding agents
Agentic workflows
Prompt & harness design
Agent evals & evaluation harnesses
Behavior & regression evals
Repository automation
Validation-driven development
CI/CD

Languages

Python
Java
Rust
JavaScript
C
SQL
Zig

Frameworks

Pandas
Django
React
TypeScript
Node.js
Flask
NumPy
Matplotlib

Databases

PostgreSQL
MySQL
SQLite

Current Work Spectrum

A snapshot of where I spend my time across technical delivery, coordination, validation, automation, and agentic product work.

Signal junction