Jayden Ontalan Capistrano Full-stack applications / applied AI

Full-stack applications for messy, real-world workflows.

I build responsive web products, Python-backed systems, and applied AI tools—from collection and planning applications to retrieval and citation systems.

View projects Résumé GitHub

Featured work

Selected application work

Focus

Usable products, data-backed workflows, applied AI

Building

  • Ahhgela Ledger active, public browsing
  • Hangout Planner production preview, no-login
  • SpineFrame production preview, private runtime
  • MFS RAG research prototype, public interface observed

Technologies

React · TypeScript · Python · APIs · PostgreSQL · self-hosted

Responsive workflowsPython-backed systemsRetrieval and citation
Web
Responsive product workflows
Backend
Python, APIs, PostgreSQL
AI and data
Retrieval, grounding, evaluation
Delivery
Public and self-hosted deployments

Selected work

Selected applications spanning collection, planning, and Python-backed systems.

All projects
Ahhgela Ledger populated collection interface with summary counts, filters, regional grouping, and sticker tiles

Data-driven web application

Ahhgela Ledger

Active Public browsing; private admin tools

A responsive collection ledger for a Pokémon sticker pack with no authoritative master catalogue. It distinguishes verified and possible entries, tracks owned and desired quantities, and supports fast region-based and mobile editing.

Highlights

  • Modelled collection state around an incomplete and evolving catalogue.
  • Refined region navigation and repeated editing workflows for desktop and mobile use.
  • Keeps verified and possible sticker entries separate so reference data does not become a verified collection claim.

Project focus

  • Incomplete catalogue modelling
  • Region-based browsing
  • Repeated edit workflows
ReactTypeScriptViteNode.js

Sole developer· Updated on Aug 2026

Open application View project

Collaborative web application

Hangout Planner

Production preview Public; no-login participation

A public planning application for collecting availability, running polls, and coordinating group decisions through shareable links without requiring participants to create accounts.

  • Supports low-friction planning links for casual events, availability, polls, follow-up questions, notes, and group-facing results.
  • Tightens the creation, participation, and results flow with grouped question cards, required-field validation, completion/share-link state, and clearer saved/unsaved cues.
  • Keeps the participant path account-free so a shared link can carry the planning workflow.
ReactTypeScriptViteNode.js

Updated on Jul 2026

Try the application View project

Python full-stack application

SpineFrame

Production preview Public login; private authenticated runtime

A self-hosted Python, Flask, and PostgreSQL application with authenticated user flows, persistent data, review and editing workflows, and end-to-end deployment ownership.

  • Builds authenticated user flows, persistent memory data, review/editing paths, and account/settings surfaces.
  • Operates as a self-hosted Flask/PostgreSQL system with documented deployment ownership.
  • Carries a 672-test validation baseline for the current production-preview build.
PythonFlaskPostgreSQLNginx

Updated on Jun 2026

Open login page View project

Applied AI spotlight

Mānoa Faculty Senate RAG

A Python retrieval and citation prototype for public Faculty Senate records, combining document extraction, metadata policy, hybrid lexical and vector retrieval, source grounding, and scoped evaluation.

  • Builds Python document-processing pipelines for local Faculty Senate PDFs and DOCX files.
  • Combines metadata policy, lexical matching, LanceDB vector retrieval, and grounded citation responses.
  • Adds structured date/action facts, response-contract checks, evaluation evidence, and current technical boundaries.

Developer tooling and systems research

Deeper workflow, validation, and investigation projects that sit behind the application work.

Project Steward

Active build

AI-assisted workflow records, guarded model calls, and explicit state decisions.

View project

Finch Standards

Active local tooling

local project standards and validation tooling

View project

Legacy Protocol Lab

Private lab

private systems investigation and evidence methodology

View project

Project Clarifier

Methodology scaffold

local-only scaffold for reducing architectural unknowns

View project

Additional applications and modelling work

Additional apps and modelling projects across vocabulary study, local forecasting, and spreadsheet-native calculation.

VocabVerse

Production preview

deployed Japanese vocabulary app

View project

Financial Forecaster

Paused

local forecasting and stress-test app

View project

Mage's Survival Calculator

Live workbook, version-bound

formula-heavy spreadsheet calculator

View project

How I work

The useful version is the one that fits the workflow, stays maintainable, and can be checked later.

  1. 01

    Understand the workflow

    Identify the actual user, source data, constraints, failure cases, and decisions the application needs to support.

  2. 02

    Build the complete path

    Connect data modelling, backend behaviour, interface design, and deployment rather than stopping at an isolated prototype.

  3. 03

    Validate and refine

    Test important states and edge cases, observe the working system, and use the results to guide further iteration.

Across the featured work

Collection, planning, Python backend, and retrieval work each test a different part of building usable systems.

  • Ahhgela Ledger responsive collection system for an incomplete and evolving sticker catalogue.
  • Hangout Planner public no-login planning app for availability, polls, notes, and group-facing results.
  • SpineFrame self-hosted Python application with backend persistence and memory-system architecture.
  • Mānoa Faculty Senate RAG retrieval and citation prototype with extraction, grounding, and evaluation evidence.

Projects, résumé, or contact.

The primary path stays short: inspect the work, check the résumé, then reach out directly.

Résumé Contact