Skip to content

Emerson Herrera Tolentino

Software Engineer · Machine Learning & Generative AI

I design and ship production-ready Generative AI systems — RAG platforms, enterprise chatbots and agentic AI — on top of 8+ years of software engineering.

About

Software Engineer with 8+ years of experience, specialized in Machine Learning and Generative Artificial Intelligence. I combine a strong background in scalable software architecture with advanced ML techniques to build intelligent solutions applied to real-world problems.

In the last two years I have focused on designing and shipping production-ready GenAI systems: enterprise chatbots, Retrieval-Augmented Generation (RAG) platforms, multimodal document ingestion, prompt engineering, LLM integration, embeddings, vector databases and Agentic AI. At Canvia I design, build and improve a multi-tenant RAG platform serving several applications of the Central Reserve Bank of Peru, including the assistant live on its Research Portal.

I am a special student at the Graduate Program in Computer Science (PPGC) of UFRGS, applying to the Master's program, and co-author of an extended abstract on vision-language models accepted at the LatinX in CV Workshop at CVPR 2026.

My goal is to build and evaluate reliable AI systems that combine information retrieval and LLMs, applied to real-world problems.

years building software for banking, telecom, health and education
8+
years building software for banking, telecom, health and education
Central Reserve Bank of Peru applications served by one RAG platform
4
Central Reserve Bank of Peru applications served by one RAG platform
F1 after fine-tuning a vision-language model (CVPR 2026 workshop)
0.19 → 0.90
F1 after fine-tuning a vision-language model (CVPR 2026 workshop)
AI Engineer Specialization, Diploma of Excellence
19/20
AI Engineer Specialization, Diploma of Excellence

Experience

  1. Oct 2025 — Present

    Analyst Programmer II · Full Stack AI Engineer / GenAI & RAG DeveloperCanvia — Client: Central Reserve Bank of Peru (BCRP)

    Designed, built and improved a multi-tenant RAG platform serving 4 of the bank's applications (BCRP Portal, Research Portal, NEXO 360 and Banknotes), including the ingestion and indexing of thousands of documents; its assistant is live on the Research Portal (investigacion.bcrp.gob.pe). Node.js gateway orchestrating client systems, Python backend with shared tools, configurable chat, search and prompt modules, multimodal ingestion and vector search on Oracle 23ai, pgvector, Pinecone and Chroma. SQL pipelines exporting research datasets to RePEc. Daily use of AI agents with human review and testing.

    • Python
    • FastAPI
    • LangChain
    • Node.js
    • Oracle 23ai
    • PostgreSQL
    • pgvector
    • OpenAI
    • AWS
    • Docker
  2. Nov 2024 — Oct 2025

    Full Stack DeveloperWay Tech — WIN Empresas / Universidad Continental

    Built academic and institutional platforms for Universidad Continental (24,000+ students across Huancayo, Arequipa, Cusco and Lima) with Angular and microservices in C# (.NET) and Java. Documented RESTful APIs with Swagger, automated financial and business processes with Node.js, and optimized SQL and NoSQL databases for high-traffic workloads.

    • Angular
    • C#
    • .NET
    • Java
    • Node.js
    • SQL Server
    • Oracle
    • PostgreSQL
    • MongoDB
  3. Sep 2024 — Present

    Full Stack & AI DeveloperCertilab S.A.C. (freelance)

    Certification system in Laravel 11 and Filament with 300+ certificates issued: admin back office, client intranet and a public QR validation portal with encrypted IDs. Corporate website with Next.js 15 and React 19. RAG and chatbot over calibration certificates with multimodal ingestion; agentic and adaptive versions with LangGraph, FastAPI and Qdrant.

    • PHP
    • Laravel
    • Filament
    • MySQL
    • AWS S3
    • Next.js
    • Python
    • LangGraph
    • Qdrant
    • Docker
  4. Mar 2021 — Aug 2024

    Systems EngineerIndra — Client: Telefónica del Perú

    Real-time processing of home modem telemetry across Peru by zone: Python ETL microservice integrating network status from several APIs, raising alerts and feeding customer management and monitoring dashboards. Distributed systems and microservices with Java Spring Boot, PHP, .NET and Python; REST/SOAP integration with JWT; database and Red Hat Linux administration; unit testing with JUnit and Mockito.

    • Java
    • Spring Boot
    • Python
    • PHP
    • .NET
    • Oracle
    • SQL Server
    • MongoDB
    • Linux
  5. Feb 2020 — Mar 2021

    IT Systems AnalystMultilab

    Transactional clinical systems with Laravel and CodeIgniter: new features, reporting, roles and permissions; database modeling and optimization; AWS (S3, EC2) and GitLab; remote and on-site technical support.

    • PHP
    • Laravel
    • CodeIgniter
    • MySQL
    • AWS
    • GitLab
  6. Nov 2017 — Jan 2020

    Software Development Area Manager · Software DeveloperMControl Energy Technology SAC

    Area manager (Jul 2019 — Jan 2020): led a team of 2 developers and planned custom software projects for service stations of the Gasocentro and Irasa groups in Lima and Ica. Software developer (2017 — 2019): custom transactional systems with C#, ASP.NET, Laravel and Vue.js on SQL Server and MySQL.

    • C#
    • ASP.NET
    • PHP
    • Laravel
    • Vue.js
    • SQL Server
    • MySQL

How I work

I use AI as a working tool every day — at Canvia and in my own projects — with the same discipline as any engineering practice: clear scope, tests and human review before every delivery.

  • Gentle Shell
  • Gentle AI
  • Engram
  • Claude Code
  • Codex
  • LangChain
  • Python
  • FastAPI
  • Coding agents with a controlled workflow

    Claude Code, Codex and Pi on Gentle Shell and Gentle AI, following ODD (Organic Driven Development): explore before changing, subagents to implement and verify, TDD and risk-based review before every commit.

  • Persistent memory

    Engram keeps architecture decisions, findings and task state across sessions, so context is never lost between days or projects.

  • Environment tailored to real projects

    Custom skills, MCP servers wired to real systems (read-only Oracle, Laravel Boost, container logs, GitLab, CodeGraph) and model selection per task.

  • Agent factory on my own VPS

    Unattended issue execution triggered from Telegram (n8n) and nightly batches, with pull requests and CI as quality gates.

Research

LatinX in Computer Vision (LXCV) Workshop · CVPR 2026 · Denver

Benchmark of Multimodal Vision-Language Models for Damage Detection

With Mariana Risco Cosavalente and Prof. Claudio Jung (UFRGS)

We evaluated CLIP, SigLIP, SigLIP2 and FuseLIP for disaster damage detection on CrisisMMD and DamageDataset, comparing contrastive fine-tuning, two-stage few-shot adaptation and multimodal fusion. Combining image and text consistently outperformed either modality alone, and fine-tuning raised SigLIP's F1 from 0.19 to 0.90.

Code

Selected case studies

  • Production RAG platform

    Multi-tenant RAG platform for BCRP

    Problem
    Several institutional applications needed reliable search and assistant experiences over thousands of research documents.
    My contribution
    At Canvia, I designed, built and improved the platform, including ingestion and indexing, shared tools, configurable chat/search/prompt modules, multimodal ingestion, and daily testing with human review.
    Architecture & stack
    Node.js gateway, Python backend, Oracle 23ai, pgvector, Pinecone, Chroma, SQL pipelines to RePEc, and multimodal document processing.
    Result
    One platform serving four BCRP applications; its assistant is live on the Research Portal.
    Read full case studyOpen Research Portal
  • Agentic & adaptive RAG

    Certilab RAG for calibration certificates

    Problem
    Calibration certificates combine structured fields and document content that users need to search, understand and validate.
    My contribution
    I designed agentic and adaptive RAG services with structured/semantic search, source citations, access control, query rewriting and answer verification.
    Architecture & stack
    LangGraph, FastAPI, Qdrant, OpenAI and MySQL; the adaptive version works over 3,848 indexed vectors from real certificates.
    Result
    A certification system with 300+ certificates issued, plus a production-oriented RAG and chatbot foundation.
    Read full case studyView project

Projects

  • E-commerce Microservices

    E-commerce platform on microservices: API gateway with authentication, product, order, payment, notification and monitoring services.

    • TypeScript
    • Node.js
    • Microservices
  • Whisper Transcription API

    Transcription API with Whisper that returns subtitle-ready JSON segments.

    • Python
    • FastAPI
    • Whisper
    • Docker
  • Posts API in Three Stacks

    The same REST API in Laravel 11 and Django REST Framework over a shared MySQL database, with a Next.js client.

    • Laravel
    • Django
    • Next.js
    • MySQL
    • Docker

Skills

AI & Machine Learning

  • Vision-language models (CLIP, SigLIP)
  • Scikit-Learn
  • PyTorch
  • LLM evaluation & LLMOps
  • Vector databases (pgvector, Pinecone, Chroma, Qdrant)
  • MCP
  • Prompt Engineering
  • OpenAI API
  • LangGraph
  • LangChain
  • RAG (agentic & adaptive)

Backend

  • Microservices & REST/SOAP APIs
  • PHP (Laravel)
  • Node.js
  • C# (.NET)
  • Java (Spring Boot)
  • Python (FastAPI)

Frontend

  • TypeScript
  • Next.js
  • React
  • Angular

Data

  • ETL pipelines
  • MongoDB
  • MySQL
  • SQL Server
  • Oracle
  • PostgreSQL

Cloud & DevOps

  • Linux
  • CI/CD (GitHub Actions)
  • Kubernetes
  • Docker
  • Azure
  • AWS

Tools

  • Jira
  • Git, GitHub & GitLab
  • Claude Code, Codex & Pi (Gentle Shell, Gentle AI, Engram)

Soft skills

  • Team leadership
  • Scrum

Education

  1. Aug 2026 — Present

    Special Student — Graduate Program in Computer Science (PPGC)

    Universidade Federal do Rio Grande do Sul (UFRGS), Brazil

    CMP263 Machine Learning (2026/2); applying to the Master's program in Computer Science.

  2. Mar 2026 — Aug 2026

    AI Engineer Specialization Program — Diploma of Excellence (19/20)

    Data Science Research Perú

    92 academic hours: RAG, agents, MCP, multimodal applications, LLM evaluation, guardrails and LLMOps. Final project: a multi-agent system applied to enterprise processes.

  3. Jun 2024 — Sep 2025

    Training Program in Microservices with Java and Spring Boot

    Lite Thinking

  4. Mar 2021 — Dec 2024

    Bachelor's Degree in Software Engineering

    Universidad Tecnológica del Perú (UTP)

    Degree conferred on February 6, 2025.

  5. Mar 2016 — Jul 2019

    Software Development Technician — 3-year technical degree

    SENATI

Certifications

  • AI Engineer Specialization — Diploma of Excellence (19/20)

    AI Engineer Specialization — Diploma of Excellence (19/20)

    Data Science Research Perú · Aug 2026

  • Natural Language Processing

    Natural Language Processing

    Data Science Research Perú · Jul 2026

  • DevOps Multicloud Specialist

    DevOps Multicloud Specialist

    Tecylab · Oct 2025

  • Angular Specialist

    Angular Specialist

    Tecylab · Jul 2025

  • Senior in NodeJS Microservices with Cloud

    Senior in NodeJS Microservices with Cloud

    Tecylab · Mar 2025

  • Senior in Spring Boot Programming on Multicloud

    Senior in Spring Boot Programming on Multicloud

    Tecylab · Nov 2024

  • Scrum Foundation Professional Certification (SFPC™)

    CertiProf · Jul 2024

Let's talk

Let's talk

I'm open to roles in Generative AI and software engineering, and to research collaborations. The fastest way to reach me is by email or LinkedIn.