José GuzmánMessage me on LinkedIn

José Guzmán

Software & AI Engineer

I build frontends, mobile apps, backends, AI agents, databases, infrastructure.

Docker · Linux · AWS · CI/CDWeb appReact · Angular · VueMobile appReact NativeBackend / APIPython · Scala · TypeScriptAI agentLangChain · LangGraph · RAGDatabasePostgreSQL · Redis · pgvector

Experience 4 yrs 4 mos

Open a role to see what I built there.

Software company I co-founded and run part-time alongside my full-time roles, building custom business systems for clients in Ecuador. I run projects end to end, from the first meeting to deployment, and share the technical lead on each one with another senior engineer.

  1. Co-built the system an engine workshop has run on since July 2025

    971 work orders and ~USD 433,000 invoiced through it. I worked on it from the contract to the Docker deploy.

  2. Co-built the operations platform a freight forwarder runs daily

    Nine people ran 324 shipments on it in its first ~2.5 months. Shipment data is pulled from PDFs with AI.

  3. Co-built a multi-company ERP with electronic tax invoicing

    ~30 modules and 175+ tables for an automotive service company. I carried more of the front end and shared the backend.

  4. Co-built a RAG checker for public procurement documents

    The legal corpus is versioned, so any past analysis can be reproduced. Each LLM call is stored with its tokens and latency.

  5. Took clients from first meeting to signed contract

    Requirements, NDAs and SLAs, and almost a year of negotiation on the largest one.

Built with
  • Python
  • FastAPI
  • PostgreSQL
  • pgvector
  • Next.js
  • React
  • React Native
  • Docker
  • GitHub Actions

AI Engineer on a platform where businesses build AI sales agents for WhatsApp, Instagram and Messenger. I focused on the agent’s reliability, and worked across the backend, web console and mobile app around it.

  1. Started the eval suite for the agent, and wrote most of it

    Each scenario runs 3 to 8 times with custom Python assertions, so I measure pass rates, not a single pass.

  2. Built the leads, calendar and resources modules from scratch

    None of them existed before I joined.

  3. Brought automated testing to a web console that had none

    24 end-to-end specs. The first one reproduced a bug before the fix went in.

  4. Tracked a payment bug down through the traces

    I had added Langfuse. The trace showed no AI call, so the fix went in the webhook, not the prompt.

  5. Replaced ~25 text-parsing hacks with typed output

    Rolled out in phases behind a flag, without stopping feature work.

  6. Cut the dashboard load from ~8.8 s to ~3.1 s

    And the inbox from three database round-trips to one, with local token verification instead of a call to the auth provider on every request.

Built with
  • Python
  • FastAPI
  • LangGraph
  • promptfoo
  • Langfuse
  • LangSmith
  • PostgreSQL
  • pgvector
  • MongoDB
  • Redis
  • RabbitMQ
  • Angular

Software consultancy serving international clients. I worked on a Python platform for ski competitions, and on front-end and QA work for other clients.

  1. Made a ski competition platform update scores live

    Server-sent events and HTMX on Django and Flask templates, which had no reactivity of their own. Until then, operators reloaded the page by hand.

  2. Built features for international and US federation rules

    What the platform needed before it could be used in sanctioned competitions.

  3. Labelled the training data for automatic jump scoring

    Jumps, scores and tricks, for a YOLO model on SageMaker. The model itself was built by others.

  4. Migrated a client’s site from Vue 2 to Vue 3 and Nuxt 4

    From Bootstrap to Tailwind, split into modular components with composables.

  5. Took QA tickets on a Java CMS for a large communications client

    Adobe Experience Manager, new to me. I tested the tickets and fixed a few of them myself.

Built with
  • Python
  • Django
  • Flask
  • htmx
  • Vue.js
  • Nuxt
  • Tailwind CSS
  • Java

Torre is a hiring platform with over a million user accounts, and EmmaLLM is its multi-agent system that talks to and screens candidates for employers. I worked on how the agent talked to candidates, and was code owner of the process layer underneath it.

  1. Cut users blocking us by more than half with a WhatsApp opt-in

    I held notifications until the candidate gave consent, instead of messaging first.

  2. Stopped the follow-ups once a candidate said they were out

    The system now noticed when someone didn’t want to finish the application, saved the reason and cut the screening reminders.

  3. Changed when candidates get asked for reference checks

    Only for jobs up to USD 50K or with compensation to agree, skipping current roles and education.

  4. Cleared ~3,500 notifications stuck waiting for a callback

    Along with other delivery bugs: an unhealthy notifications queue and mixed English/Spanish copies.

  5. Built agent features in Langchain4j on a Scala backend

    Inside the team’s five-agent pipeline, connected to the EmmaBPM flows that moved candidates through their applications.

Built with
  • Scala
  • Play Framework
  • Langchain4j
  • Pinecone
  • Kotlin
  • Node.js
  • Vue.js
  • Nuxt
  • ScalaTest

A startup building AI tools that generate advertising content: copy, images and banners. I worked on the front end, both cleaning up its code and building the chat and payment flows on top of it.

  1. Cleaned up a front end full of duplicated logic

    Moved the logic into hooks and server state into React Query, instead of repeating it in every component.

  2. Wired the real-time chat to an OpenAI-backed service

    The front end side of the socket.io connection was mine. Users asked the chat for ad copy.

  3. Built the PayPal credits flow

    Users bought credits with the PayPal JavaScript SDK and spent them generating images or copy.

  4. Merged a rebrand branch that had drifted from our work

    Another freelancer redesigned the whole app in one branch; I did the merge and resolved the conflicts.

Built with
  • React
  • Next.js
  • TypeScript
  • Tailwind CSS
  • React Query
  • Storybook
  • Socket.io
  • PayPal

Torre is a hiring platform with over a million user accounts and about 50 engineers. My first engineering role: an internship, but full-time, mostly on the Scala backend of the chatbot that contacts applicants.

  1. Fixed an ORM bug that lost tens of thousands of notifications

    Pulling too many at once broke Slick’s limit. I moved retrieval onto a queue of IDs and refactored the chatbot flow around it.

  2. Turned process diagrams into features across Scala services

    Each feature started as a change to a BPM: I found the new condition in the diagram, then where it lived in the code.

  3. Wrote the SQL metrics for the features I shipped

    Applicants per job opportunity, notifications sent and failed, candidates by channel, on Metabase. The squad read them in the daily.

  4. Fixed Vue.js bugs for three other squads between my own tickets

    The operators’ console, the job pages candidates saw, and the recruiters’ hiring pipeline.

Built with
  • Scala
  • Play Framework
  • Slick
  • Vue.js
  • SQL
  • Metabase
  • Twilio

Internship at a Quito software studio that builds web and mobile apps for its clients. I worked on six projects in ten months; these are the ones I took furthest.

  1. Rebuilt a Miami car wash’s website from a designer’s mockup

    I owned it end to end in WordPress, SEO included. It stayed live years after handoff.

  2. Shipped a sports short-video app to App Store and Google Play

    I did the redesign and ran both store submissions myself.

  3. Built screens for a KPI dashboard for a company’s owners

    React and Next.js with react-hook-form. Some real KPIs were already showing on it.

  4. Took features from the Product Owner to production

    Through pull requests in GitLab, on React, React Native and Firebase apps.

Built with
  • React
  • React Native
  • TypeScript
  • JavaScript
  • Firebase
  • Next.js
  • WordPress
  • GitLab

Activity

3,704 contributions in the last yeargithub.com/joseguzmann ↗
  • 2026 3,621
  • 2025 564
  • 2024 598
  • 2023 142
  • 2022 139

3,440 of them are in private repositories — GitHub counts them without showing them. Updated 2026-11-01.

Live systems I built

Private client systems. Happy to show you any of them on a call.

Side projects

Things I build for myself.

Skills

  • AI & LLM systems

    • LangGraph
    • LangChain
    • Langchain4j
    • pgvector
    • Pinecone
    • promptfoo
    • Langfuse
    • LangSmith
  • Languages

    • Python
    • TypeScript
    • JavaScript
    • Scala
    • Java
    • SQL
  • Backend

    • FastAPI
    • Django
    • Flask
    • Play Framework
    • Node.js
  • Frontend

    • React
    • Next.js
    • Vue.js
    • Angular
    • React Native
  • Data & infra

    • PostgreSQL
    • MongoDB
    • Redis
    • RabbitMQ
    • Docker
    • AWS
    • GitHub Actions

Escuela Politécnica NacionalB.Eng. in Computer Science · Escuela Politécnica Nacional

Spanish NativeEnglish C1

open to remote · UTC−5

Need to build something?

A role to fill or a system to build. Tell me about it.