Freelance developer buildingweb apps, APIs & automationfor teams that need reliable delivery.

A calm, local-first planner that keeps goals, tasks, notes, boards, and timelines useful even when the network disappears.

Problem

Planning tools often fragment work across disconnected lists and become unreliable when a connection or a hosted service is unavailable.

Constraint

The product must work offline and sync between devices without turning private goals and notes into readable server data.

Decision

Use IndexedDB as the source of truth, encrypt sync records on-device, and keep push payloads deliberately redacted.

Outcome

A calm installable planner works locally first, while optional sync preserves privacy without changing the daily workflow.

A reader-first interactive-fiction platform that pairs curated story worlds with controlled AI turns, persistent memory, and transparent usage.

Problem

AI fiction can feel limitless but quickly loses authorship, continuity, and a clear reason for readers to trust the next turn.

Constraint

The platform needs creative freedom without hidden costs, disposable memory, or a generic chat-first reading experience.

Decision

Anchor play in curated story packs, persist memory around reader choices, and use one visible Ink cost table for turns, retries, images, and context.

Outcome

Lorebound keeps the reader inside authored worlds while making AI capability, continuity, and usage understandable.

A motorcycle command center for fuel economy, maintenance, tires, documents, parts, and cost intelligence.

Problem

Riders track fuel, maintenance, documents, tires, and costs across scattered notes, apps, and spreadsheets.

Constraint

The product needs app-like feedback without making the operational dashboard difficult to maintain or deploy.

Decision

Keep Django in charge of the domain and server-rendered UX, then use HTMX for focused interactions and Docker for a repeatable runtime.

Outcome

One operating surface now connects upkeep, cost visibility, reminders, and a future profitability workflow for professional riders.

A daily wildfire ignition-risk modeling system for Goias, Brazil, built around versioned data, climate signals, and reproducible evaluation.

Problem

Daily wildfire ignition risk needs comparable results across climate, dryness, vegetation, and historical conditions.

Constraint

A model improvement is not useful if its data, assumptions, and evaluation cannot be reproduced or explained.

Decision

Treat the research pipeline as a product with schema contracts, manifests, experiment profiles, registry entries, and diagnostics.

Outcome

The materialized climate and MODIS line improved PR-AUC from 0.407 to 0.456 while preserving an auditable path forward.

An innovation-hub portal for UniRV's entrepreneurship ecosystem, bringing incubated startups, events, and institutional services into one place.

Problem

UniRV's innovation ecosystem had no single entry point for startups, mentors, partners, and the public to find incubation programs, events, and institutional information.

Constraint

The portal had to serve very different audiences — applicants, partners, and compliance/transparency requirements — without becoming a slow, hard-to-maintain institutional CMS.

Decision

Built a server-rendered Django application with HTMX for focused interactivity and Alpine.js for lightweight client-side behavior, shipped as containers behind Docker Swarm and managed through Portainer.

Outcome

AgroHub UniRV now centralizes the startup directory, events calendar, service listings, and institutional access channels in a portal that stays fast and simple to operate.

A city-level innovation ecosystem site for Rio Verde, mapping local hubs, incubators, and programs along an interactive geospatial trail.

Problem

Rio Verde's innovation ecosystem — hubs, incubators, universities, and programs — had no shared visual way to show where the city's innovation actors are and how they connect.

Constraint

The site needed an interactive geospatial layer without the latency and operational cost of a heavy traditional backend, while staying easy to update.

Decision

Built the frontend in Next.js with Mapbox, MapTiler, and deck.gl for the interactive trail, backed by Neon Postgres accessed through Cloudflare Hyperdrive for low-latency edge queries, with Neon Auth handling access control.

Outcome

Inova Rio Verde presents the city's innovation ecosystem as an explorable geospatial trail instead of a static list of institutions.