EPEduardo Polanco
Open to work LA --:--:-- Uptime --

build 2026.10 · Los Angeles, CA · English / Español

Eduardo Alfredo Polanco

AI Reliability & Automation Engineer

I build production software with AI, and I catch what it gets wrong before it reaches a live system.

My workflow is human-in-the-loop and multi-model. Cursor, Claude, and Gemini build and cross-check each other, and I review every change before it ships. That review step has caught real problems before they reached production, from automations that would have flooded a live CRM to security gaps in internal tools.

Before tech, I spent ten years taking care of animals, first at a grooming shop and then at a vet clinic. That work taught me to stay calm under pressure and double-check everything. I still work that way.

122call tasks stopped before launch 39security issues found, 10 high 9 mobug solved in my first week

Caught in review

Tap any catch for the story
  • 122call tasks stoppedHIGHOne command would have created 122 real call tasks in a live CRM.
    Situation
    A Python script was about to run a 15-touch sales sequence against the company CRM.
    What I did
    Code-reviewed it on my first call with the developer. It was pointed at the live CRM, not a test copy, so one command would have dropped 122 real call tasks on the sales team. I flagged 6 issues in total.
    Result
    All 6 were fixed the same night. The launch became a 5-lead pilot, with the cadence approved by sales leadership first.
  • 39issues found, 10 highHIGHA regular account could grant itself executive access.
    Situation
    After seeing my sales-script review, the developer of an internal operations app (TanStack + Supabase) asked me to audit it.
    What I did
    Ran a backend audit and then a front-end audit. I checked every finding against the actual code before logging it with a severity, due date, and dependencies.
    Result
    39 verified issues, 10 high severity, including AI features anyone could use without logging in, an assistant account that could promote itself to executive access, and private files open to every user. I also proposed a menu redesign cutting about 40 links to 6 sections.
  • 4gaps flaggedHIGHA "sandbox" CRM was wired to live sales data.
    Situation
    I took over backend and security for a simplified CRM built on top of the company’s main one.
    What I did
    Reviewed the existing build before adding anything, and found it was connected to the production CRM instead of a sandbox, so any test could write to real sales records.
    Result
    Flagged 4 gaps, starting with the live connection, and recommended a lower-cost database to keep spending down until the product earns revenue.
  • 13findings, verifiedVERIFYChecked an AI’s security audit against the real code and caught a false alarm.
    Situation
    I was given access to a production app’s code and used AI to help audit it.
    What I did
    Locked my setup first: read-only token, pushes blocked, AI in ask-only mode. Then I checked each AI finding against the code instead of trusting it, and caught that a "missing" security step had been left out on purpose.
    Result
    A 5-page audit (6 high, 3 medium, 4 low) and a 4-phase plan, sent to leadership the same day. Leadership replied within minutes, and I now own the backend plan.
  • 9 moblocker, solved in week 1MEDA bug board had been stuck for 9 months. I diagnosed it in my first week.
    Situation
    A startup’s bug-tracking board had been stuck for 9 months. Nothing could stay marked as done, and nobody knew why.
    What I did
    I had no production access yet, so I treated it as a read-only investigation. I went through every automation rule and found one that sent any task moved to Done back to review with no approval check. The timing matched exactly when it broke.
    Result
    Posted the root cause and two fix options without changing anything. I also showed the 6 "open" bugs were finished fixes waiting on sign-off.
  • 44possible duplicate dealsINFOMapped every automation rule in a live CRM in about 2 hours.
    Situation
    Operations asked for a full picture of what the CRM automates.
    What I did
    Checked 11 areas across every deal pipeline and lead flow, including rules hidden in the workflow designer, without changing anything.
    Result
    Found zero triggers, zero automated client emails, three pipelines with no automation, and 44 possible duplicate deals. Recommendations were delivered 4 days early.
  • 5/5test scenarios passedPASSPricing calculator matched the official card in 5 of 5 test scenarios.
    Situation
    Sales asked if a deal could have checkboxes that change the total.
    What I did
    Built it the same day: 8 credit options plus a pick-one case study, with the weekly fee calculated from the official pricing card.
    Result
    All 5 test scenarios matched the card, including everything selected at $200. Sales confirmed the default that evening.

git log

main · newest first
  1. HEADa3f9c21Full Stack EngineerThe JobBridge, Inc. · contract · remote

    Full stack work across the company's apps, including The JobBridge platform and AnchorChart Pro: backend, database, integrations, and App Store submissions. In my first week I diagnosed a bug that had blocked the QA board for 9 months, and two weeks in I was helping screen and interview new hires.

  2. HEAD7be04d8Automation & Integrations SpecialistLoyalty Untapped · contract · remote

    First hire in a new role, building the CRM automation layer in Bitrix24 from the ground up. My contract e-sign workflow became the foundation of the sales-to-invoice pipeline: CRM → contract → QuickBooks invoice. Now I own 2 of the Tech team's 4 Q4 priorities and the backend plan for the company's app.

  3. HEADe19c4f0Front-End Developer Micro-CredentialWestern Governors University
  4. HEAD52d7a3bStarted B.S. Software EngineeringWestern Governors University · in progress

    Software security and testing, design and QA, Android development, Java frameworks, back-end programming.

  5. HEADc8a61e2Project Repository MaintainerWest Los Angeles College · Culver City

    Led a team GitHub repo with Agile workflows: reviewed contributions, coordinated pull requests, resolved merge conflicts, and wrote the docs the team worked from.

  6. HEAD0f3b9d7A.A. Computer Science & IT, Dean's Honor RollWest Los Angeles College · + Google IT Support Certificate
  7. HEAD9d24e6aIT End User Device SupportPer Scholas / TEKsystems
  8. HEAD4c7710emerge: animal-care → software

    Left the clinic in mid 2022 and started Computer Science at West LA College, bringing ten years of checking everything twice with me.

  9. HEAD7e2b40aA.S.-T. Administration of JusticeWest Los Angeles College

    Studied criminal justice, law, and investigations while working at the clinic. That's where I started thinking about access, evidence, and accountability, which shows up in my security audits today.

  10. HEADb63f1c9Veterinary TechnicianOverland Veterinary Clinic · Los Angeles

    Started as a Certified Veterinary Assistant, promoted to Technician Manager Assistant in 6 months and to technician-level duties within a year. Vaccines, IV catheters, blood draws, anesthesia and vitals monitoring for dogs, cats, and exotics.

  11. HEAD36ae08dCertified Veterinary Assistant, Magna Cum LaudeAmerican College of Healthcare & Technology · top 2 in class
  12. HEADd0e5b72Promoted to ManagerWags 2 Whiskers · Los Angeles

    Ran daily operations, scheduling, staff, client communication, and intake.

  13. HEAD1a0c3e5Initial commit: BatherWags 2 Whiskers · Los Angeles

Track record

Fall 2026
2 of 4Top priorities on the Tech team this quarter are mine to deliver: the sales CRM and client-onboarding automation.
44 testsWritten in the first 2 hours of a new training-system build, with security set up before any features.
16 minTo unblock the team's top priority with one clear message, instead of waiting a week for a meeting.
Read-onlyHow I review company code. Before opening it, I locked my setup so neither I nor the AI could change a single line.
  • My blueprint was picked. Two of us designed an onboarding training and quiz system separately, and my 11-page MVP plan was chosen as the base to build from.
  • Hiring, two weeks in. Asked to help screen resumes and interview candidates for the team.
  • Pricing strategy. Reviewed a SaaS product's costs across 16 categories and delivered recommendations on pricing, conversion, and school and district sales.

Projects

Status · Stack

VERAVerify · Explain · Repair · Audit

Rebuilding

Drop in the code you're auditing and get the real errors and security issues fast. VERA writes a full fix brief into the project folder so any IDE's AI picks it up, plus a short plain-English card for the human. Checks for exposed secrets, OWASP Top 10 patterns, and known-vulnerable packages.

Python · Claude API · Gemini API · OSV.dev · Semgrep · Bandit · gitleaks

Yip & HowlCommunity wildlife & pet alerts

Live

Bilingual app for South LA to report wildlife sightings, lost pets, and found animals, no account needed. Built solo in about a month. An automated news pipeline summarizes local animal news with Gemini and keeps working through a fallback, and a request firewall auto-bans exploit probes and emails me.

Flask · PostgreSQL · React Native · Sentry · yipandhowl.app ↗

OlympusLocal-first multi-agent AI IDE

In development

Orchestrates specialized AI sub-agents through development workflows, with a secure API gateway, fallback routing, system constraints, and secure memory for predictable agent behavior.

Python · FastAPI · Claude API · Docker

isitRealAI image detector

Working prototype

Gives an "AI or Real?" verdict with a confidence score and explains why in plain English. Checks images with two detectors, flags when they disagree, names the likely generator, and deletes every upload after the scan. Video scanning is built and in final testing.

Python · Flask · Sightengine · Hive · Gemini

VetappPlatform for veterinary professionals

Private

Designed from my own clinical experience. In active development.

TypeScript
Smart Temp ToolThermal data analysis tool delivered to the UNLV Data Analysis Group.
Vacation SchedulerNative Android app, shipped as a signed release APK.
KoostosOffline desktop network scanner for local diagnostics.

Toolkit

What I work with
AI
Claude API, Gemini API, multi-model development, multi-agent orchestration, prompt engineering, rule-based fallbacks
Reliability
Human-in-the-loop review, OWASP Top 10, Semgrep, Bandit, gitleaks, pip-audit / OSV.dev, CSRF and rate limiting, Sentry
Backend
Python, FastAPI, Flask, TypeScript, PostgreSQL (Neon, Supabase), SQLAlchemy, REST, JWT
Frontend
React Native, Expo, Angular, TanStack, Android (Java), PWAs
Automation
Bitrix24, Make, Airtable, QuickBooks, Asana
Infra
GCP, Docker, Nginx, Gunicorn, Linux, Git, GitHub, GitLab

Got a system that needs a second set of eyes?

I'm open to full-time and contract work in AI reliability, automation, and full stack engineering.

eduardo@eapolanco.org