FIG. 01 · Robert Vassallo

i ship products, and reverse-engineer the rest.

hi, i'm robert vassallo, 19. spent this summer shipping agentic systems into production at daisy, now back at uni in leeds. been building since 12 ↳: ios apps on the store, hackathon wins, paid reverse-engineering of commercial software.

Malta → Leeds Open to '27 internships GitHub LinkedIn

SWE intern @ a startup backed by Stripe's founders

6× hackathon award winner

£3k Spark grant

Best student company nationally @ 17

Paid to crack commercial code protection @ 15

FIG. 01.A · summer '26

Robert pitching Nautilo solo to a full room at the EUDIS Defence Hackathon, Portugal

FIG. 01.B · EUDIS '26

FIG. 02 · awards

trophy cabinet.

2026 · Oct

EUDIS Defence Hackathon · Portugal

2nd overall

€3,500

nautilo · mothership · EU Defence Innovation Scheme

2026 · Jun

NVIDIA Hack for Impact

1st

6mo ElevenLabs Scaleworth $299/mo

argus · elevenlabs challenge · local on dgx spark

2026 · May

AI in the Box hackathon

Prize

£250

concordia · multi-agent runtime

2026 · Feb

UCL × PGIM Real Estate hackathon

2nd

£1,500

ghostplot · embodied carbon

2026 · Jan

Hyperliquid × Encode

2nd + 3rd

£2,250

hypergate · li.fi track + community

2025 · Nov

DurHack 2025 · Marshall Wace

1st · track

iPad + Pen + duffel

bridging worlds · 1 of 67 teams

2025

Spark grant

Selected

£3,000

biteswipe · 150+ applicants

2024

JA Europe · Malta finals

Best co.

€2,600 profit

our company, remind ↗ · 400 sales · age 17

2024

JA ISCF · Latvia (intl)

Best sales team

vs 40 teams

team award, with remind · times of malta ↗

Team Nautilo on stage with the second-place cheque at the EUDIS Defence Hackathon 2026

Nautilo · EUDIS '26

ARGUS · NVIDIA '26

DurHack · Nov '25

the work ↓

Part 01 / 02

WORK

production software, for real companies. an internship at daisy, summer '26, and a paid build for chaos, sept '26.

FIG. 03 · Daisy

this summer, i shipped to production.

ten weeks at daisy, a dublin startup backed by stripe's founders. their prediction platform manages billions of pounds of produce a year. i worked across the front end and the AI, and three things i built are in production.

the first thing every user sees is the assistant i proposed and built, on every customer's front page. it already knows what's on screen, at zero inference cost, so you just ask. it answers from the real data and shows every step it took.

Role: swe intern · front end + AI
Stack: typescript · next.js · llm agents
Shipped: 3 systems to production
Impact: 8.3× faster onboarding
Reach: 100% of clients
Date: jun to aug 2026

IN PRODUCTION 3 · systems shipped · summer '26

What I built

the assistant i built, open beside a supply chart on daisy's landing screen

daisy's landing screen. the assistant on the right is mine. figures redacted.

FIG. 03.A · grounding

a numeric grounding system for their agent.

Problem: the agent could state numbers that weren't in the data.
Fix: it now points at numbers instead of writing them. each one is traced to its source when the page renders.
Result: anything it can't trace gets flagged. hover any figure to see where it came from.

an answer where each figure is underlined and linked to the data point it came from

Hover any figure · see its source

FIG. 03.B · the front end

i unified their UI.

8.3×Faster to onboard
a new company

Problem: every customer's screens were their own code, so one change meant editing every copy by hand.
Fix: i audited the debt, proposed a rebuild, and rewrote the entire front end solo: 27 shared building blocks, one settings file per customer.
Result: it's now the team's architecture, deployed to 100% of clients.

Before · a copy per customer

customer-a/

Dashboard.tsx · Charts.tsx
Filters.tsx · Assistant.tsx

customer-b/

Dashboard.tsx · Charts.tsx
Filters.tsx · Assistant.tsx

customer-c/ …

same screens, rewritten again

After · shared blocks + one file each

blocks/ · 27 primitives

ChartTableFiltersCardAssistantLayout

customer-a.ts

{
  pages: ["supply",
          "forecast"],
  metrics: [...],
}

customer-b.ts

{ ... }

new.ts

{ ... }

The shape of it

a sketch of the structure, not daisy's actual code.

FIG. 04 · Chaos · freelance · Sep '26

i ported chaos from iOS to native android.

10Days · paid
commission

Kotlin · Compose · RuntimeShader

Brief: a native android version of the chaos iOS app, matching its behaviour and layout.
Built: the whole app, natively, in kotlin and jetpack compose.
Hard part: apple's liquid glass, on android, without dropping frames.

Part 02 / 02

PERSONAL
PROJECTS

hackathons, side projects, and things i built because i wanted to.

FIG. 05 · ARGUS · hackathon · Jun '26

the operating system for city traffic.

one nvidia dgx spark, everything local. it reads ~780 live traffic cameras, counts vehicles on-device with yolo11x, and flags a congestion cascade about 12 minutes before the official feed does. then it waits: an operator approves the fix, the bus reroutes on the map, and the system verifies the outcome and remembers the shift.

i built the entire ui, the human-in-the-loop decision flow, and the elevenlabs voice agent with the session-memory layer that took 1st in the elevenlabs challenge at nvidia hack for impact, london.

Stack: dgx spark · nemotron 70b · yolo11x · elevenlabs
Team: with alex pivovarov · jeffrey chang
Won: 1st · elevenlabs challenge
Status: shipped · all-local inference
Date: jun 2026

NVIDIA Hack for Impact 1st · ElevenLabs Challenge

Closed-loop · DGX Spark

Perception · YOLO11x

FIG. 06 · BiteSwipe · own app · Jan '26

biteswipe ships a real ios app to your pocket.

an end-to-end cooking os, designed and built solo. swipe recipes, plan a week, auto-generate the shopping list, import any recipe url. flutter front-end, supabase back-end, on the app store. funded in part by a £3k spark grant against 150+ applicants.

Stack: flutter · supabase · app store
Team: solo · front to back
Won: £3k spark grant
Status: shipped · on the app store
Date: jan 2026

On the App Store

App Store

FIG. 07 · Open Water · side project · May '26

at thirteen i used gerstner waves. came back at nineteen to take it further.

at thirteen i approximated a sea with stacked sine waves. this is the version i couldn't build then: tessendorf's fft, the method the film studios use, rebuilding the surface from a wave spectrum every frame. it finds where crests fold and crusts them with foam, and the boat samples the surface to float, pitching and rolling into the swell. three.js, on the gpu. same problem, years of math later.

Method: tessendorf ifft · jacobian-determinant foam · multi-point buoyancy
Stack: three.js · webgl · gpu
Lineage: gerstner @ 13 → fft @ 19
Status: hobby project
Date: may 2026

Real-time · WebGL

Try in browser

FIG. 08 · PyArmor · paid disclosure · Feb '23

i cracked a commercial code-protector at fifteen.

recovered a pyarmor-protected app's full source by patching cpython's marshal loader. the app's owner paid for the write-up.

i ship products, and reverse-engineer the rest.

trophy cabinet.

this summer, i shipped to production.

a numeric grounding system for their agent.

i unified their UI.

i ported chaos from iOS to native android.

the operating system for city traffic.

biteswipe ships a real ios app to your pocket.

at thirteen i used gerstner waves. came back at nineteen to take it further.

concordia runs multi-agent minecraft inside the browser.

i cracked a commercial code-protector at fifteen.

two more, shipped.

proof i could ship, not just pitch.

where the build is the interface.

compound does user testing by simulating their brains.

hot/cold guides blind hands toward objects.

i've been at this since twelve.

on the side.

routing humans out of buildings on fire.