ESEN
Back to the portfolio

featured project

Spark Match

Vocational guidance copilot powered by Generative AI

Capstone · UNI 2026In active development

A platform that helps students discover their career path by talking to an intelligent agent: it assesses their vocational profile, computes affinity with careers and builds action plans with real resources — not a chatbot, but an agent with multi-step reasoning, memory and delegation to subagents.

degree programmes analysed
550+
institutions
1,000+
regions of Peru
25

Official data from Ponte en Carrera · MINEDU Peru

Final presentation for the Generative AI & MLOps specialization at UNI, with the platform running.
  • Deep agent with 3 specialized subagents and student profile memory (langmem), grounded in Holland's RIASEC vocational model.
  • Serverless backend with DDD and event-driven architecture on Lambda and EventBridge; Angular 22 frontend with Signals and i18n.
  • Multi-environment infrastructure (dev/prod) with pure Terraform and OIDC-based CI/CD — zero access keys — plus quality gates and security scanning.
  • Real MLOps: datasets versioned with DVC, experiments tracked with MLflow and an academic paper with an automated LaTeX build.

The platform

The agent delegates to specialised subagents and surfaces every step as it reasons.
Each programme is matched against official MINEDU figures: admission rate, length, annual cost and monthly starting salary.
The report is exported as a PDF detailing every recommended programme.
Automated evaluation in LangSmith: custom evaluators and model comparison.

My role: I led the design and construction of the platform end to end — infrastructure, CI/CD, backend, frontend and the deep agent (400+ commits) — alongside teammates who supported the academic research.

GitHub organizationDeep agentInfrastructure