Team & credits

Builders behind Playground, Configure, and Sandbox

Real people, real scores, harmony you can understand and trust.

We're building notation tools with care: clear, rule-grounded algorithms (the same Glass Box idea as the product) and a calm interface that respects the musician at the desk. Below you'll meet us separately. Near the bottom we share how we fit as a pair, then the product stack and links. In a hurry? Collapse a card to skim, then open one when a story pulls you in.

Pick a builder, or just scroll and wander

Dulf Vincent Genis

Lead Fullstack · Backend & engine

Dulf carries the path from upload to generated score: intake, symbolic pipelines, the SATB runtime, export glue, and the APIs that keep the editor and engine in sync. He spends his deepest engineering time in Cursor, in long file-grounded passes where LangChain-style server routes and the Theory Inspector stay tied to real tests and observable behavior, not vibes. He cares about theory-shaped correctness, tight boundaries around the solver, and kindness to future-you through regression gates, reviews, and refactors that won't embarrass the project next month, so players and teachers can trust what ships.

Engine & APIs

SATB generation
Symbolic score pipelines
Route handlers
Performance budgets
Docker / deploy

Theory in software

Voicing rules
Solver tuning
Score validation
Constraint design

Shipping with care

Test & API gates
LangChain inspector routes
Cursor-deep refactors
Contract-clear integrations

Born in the Philippines and raised in the U.S., he studied Information Sciences & Data Science (Spanish minor) at UIUC. Faith, family, and close friends keep him grounded. He sings and plays violin and piano, often in service of community on mission trips, at local gatherings, or with a string quartet learning a new chart together.

“AI makes directions infinite—so choosing what actually matters is the skill. Professor Huang told us to pick ‘personal best interest’; that’s the only reason HarmonyForge exists…”

Shivam Patel

Lead Fullstack · Frontend & product craft

Shivam shapes what musicians actually touch: the Next.js app, Document and Sandbox flows, RiffScore integration, Theory Inspector chrome, and the visual language that keeps dense scores legible. His longest creative loops run in Claude around interaction, motion, and copy, while Figma is where layouts prove they deserve to become components. He sweats notation-first UX, calm state architecture, responsive rhythm, and the quiet tactile detail in controls (the kind of thoughtfulness you notice in a well-made door handle), so a heavy score still feels approachable instead of hostile.

Product & UI

Next.js
Design tokens
Responsive layout
Motion & polish
Accessible patterns

Notation surface

RiffScore bridge
Zustand state
Playback UX
Export / print chrome

Design × code

Figma → implementation
Claude-forward UI passes
Microcopy sweeps
Pre-ship polish rounds

Shivam is a UX-minded builder (UIUC Information Sciences; product design internship at Microsoft). Projects like GeoGroove and mentoring student builders keep him close to real people learning to ship their first good ideas.

“The gap to learn these tools is essentially zero—what changed was my mindset. Professor Huang pushed us to use AI in daily life, not only in code…”

How we work together

We're distinct on purpose. Dulf pulls the center of gravity toward the engine, intake, SATB runtime, exports, and the APIs that have to be correct. Shivam pulls it toward the product surface, notation UX, and the feel of every screen a musician actually lives in. We hold the same bar for honesty and Glass Box clarity. Different home bases mean neither of us has to be a generalist at everything.

We still complement each other constantly. Dulf's day-to-day desk is Cursor. Shivam's is Claude. We meet in the middle with small, boring artifacts that save arguments, like repro steps, API sketches, acceptance notes, screen recordings, and failing cases. We share the same context engineering habits: tight prompts, curated excerpts, wiring assistants back to real files when answers need to stay grounded, and spec-driven development when it keeps acceptance and scope legible before implementation. We share the same instinct to do a short, source-first read together when neither of us wants to guess. When material sprawls, we also push RAG-style retrieval and state-of-the-art models toward what they're built for. NotebookLM and Google's AI-assisted research tooling are in that rotation. They are not load-bearing infrastructure for the product. They help us get citation-friendly, source-tied reads without pretending the tools think for us. None of that replaces judgment. It speeds alignment so the solver and the interface don't drift apart.

Product stack

What runs in HarmonyForge today. Hover or focus a tag for a one-line note on how we use it.

App & interface

Notation & sound

Inspector, safety & ship