Live · AI sommelierRecommendedSpeyside 12 YearSingle malt
Hi! Tell me what you usually drink and I'll find you something new.
01 · Video avatar
AI-powered whiskey sommelier
- Next.js
- React
- OpenAI
- HeyGen
Clients
Built for household names.
Clerq · Search
500M
patent and non-patent documents searchable on the AI platform I work on
Clerq · Customers
2,000+
professionals at top law firms and innovative companies use it
Jam Shed · AR
1.68M
bottles carried my AR prize game, which lifted sales
Clerq · Rendering
10,000+
AI data records processed and rendered without lag
Spotlight · Minor Figures · built by me
The award-winning game that put a carton on the run.
I architected and built the front end of Notorious Penny, the game that turned an oat-milk carton into a fugitive. Players build it a disguise, then share it or play to win, all running inside the brand’s Shopify store.
Rap sheet · what I built
- Drag-and-drop disguise builder
- Two routes in: play for fun, or play to win with receipt validation
- GSAP animation on every interaction
- Snapshot your disguise and download it
- Light enough to live inside a Shopify store
01 · Brands I've built for
Interfaces for millions of people.
Hover to scan · select to open the case file →
multi-stage PIN entryopen →
WCAG-compliant micrositeopen →
× Netflix prize drawopen →
award-winning gameopen →
AR prize wheelopen →
physics Plinko gameopen →
live voting chartopen →
digital couponsopen →
timed barcode revealopen →
Lottie peel-and-revealopen →
legacy rebuildopen →
reverse-proxy integrationopen →
photo proof-of-purchaseopen →
email-verified daily drawopen →
mobile-first, SMS 2FAopen →
multi-region instant winopen →
giveaway rebuildopen →
age-gated prize drawopen →
code + receipt entryopen →
animated bottle carouselopen →
weather-API rebateopen →
GSAP drag-and-revealopen →
multi-path instant winopen →
confetti claim flowopen →
Spotlight · Monster Energy · built by me
A legacy site, brought back to life.
Monster’s Claw Club rewards site wouldn’t install, wouldn’t build and had a broken layout. I took it over and got it shipping again.
Before:Dependency tree that wouldn't install
After:Pinned versions, clean repeatable installs
Before:Native node-gyp build failing
After:Toolchain configured, native bindings compiling
Before:Dead dependencies and a tangled Webpack config
After:Leaner dependencies, simpler config
Before:Footer colliding with the carousel
After:Footer and carousel each keep their own space
Build passing · deployed
02 · AI interfaces · client work
Interfaces that talk back.
Conversational AI built for real rooms: a pitch meeting and a shop floor. Hover the numbers to see what I built.
Live · AI sommelier
Ask me anything
What goes with roast lamb?02 · In-store tablet
Virtual wine assistant
- Next.js
- iPad UI/UX
- AI chat
Event showcase
Selected for the House of Accolade Shopper event in London's Leake Street tunnels, where industry experts tried it live.
View LinkedIn post ↗03 · Ask the portfolio
Building interactive instruments to make complex AI systems legible.
Software engineer at Clerq, an AI patent-intelligence platform. Previously, interfaces for PepsiCo, Coca-Cola, LEGO × Netflix and Mars.
Ask the map
● 5 sentences lit on the map1 project
I build Clerq's AI patent-intelligence platform end to end: a vector-matching engine for semantic search, 3D visualisations of the search space, and a rendering pipeline that keeps 10,000+ AI records smooth. I work across the SaaS and agentic teams.
Clerq
FIG. 00 · every sentence about my workn = 171 · all-MiniLM-L6-v2 · UMAP
Drag to orbit · hover a point to read it · click to open its project
04 · The lab · personal projects
Things I build to understand AI.
Each one takes an idea from machine learning research and turns it into something you can play with.
FIG. 02
LossLandscapes
Why some networks train and others stall: morphing 3D loss surfaces and optimiser paths, from Li et al. (2018).
- React Three Fiber
- GLSL
- GSAP
- PyTorch
FIG. 03
Neural Manifold Explorer
5,000 embeddings from MNIST, CIFAR-10 and 20 Newsgroups in 3D. Orbit, inspect neighbours, add physics.
- React Three Fiber
- UMAP
- Python
FIG. 04
Mech Interp Lab
Inside a transformer: circuit ablation, attention patterns and the logit lens, modelled on GPT-2 small.
- React Three Fiber
- D3
- Framer Motion
FIG. 05
Diffusion Denoising Simulator
Scrub from pure noise to an image and watch a diffusion model's estimate sharpen, step by step.
- Canvas
- Python
- Zustand
FIG. 06
Bellman Policy Playground
Train Q-learning, SARSA and Monte Carlo agents on a grid world you design, and watch the policy form.
- Canvas
- React 19
- Reinforcement learning
05 · Capabilities
Model card: Seline Davies.
type: AI engineer · base: London · ● in production
- Intended use
- Building AI products that people can understand, trust and act on.
- Training data
- Philosophy at King's College London → systems → full-stack → campaigns at scale → AI at Clerq.
- Known limitations
- Will always ask why before building. Cannot resist a good visualisation.
Capabilities
01
Semantic search and retrieval
Vector databases and semantic search across 500M patent and research documents.
- Python
- Vector databases
- Embeddings
02
Agentic AI products
I move between Clerq's SaaS and agentic teams, taking AI features from model output to the interface.
- Python
- TypeScript
- LLMs
- Agents
03
On call for production
On the rota for a platform used by top law firms and innovative companies, from first alert to root cause.
- Incident response
- Datadog
- Sentry
04
Testing and observability
Tests and monitoring that catch problems before customers do, and tell us why when they don't.
- Testing
- Datadog
- Sentry
05
Interfaces people trust
Designing how people read and question model output, from Figma to accessible, WCAG-compliant code.
- React
- Next.js
- Figma
- WCAG
06
Client need to delivery plan
Turning what clients need into scoped plans, then shipping them. Campaigns for PepsiCo, Coca-Cola and LEGO.
- Discovery
- Planning
- Delivery
07
Making models legible
Visualising what happens inside AI systems, like the embedding map further up this page.
- WebGL
- Three.js
- UMAP
08 · Inference
Want to see how it performs on your problem?
Get in touchLinkedIn
06 · Design, start to finish
Good design makes complex things feel simple.
I design and build: wireframes, prototypes and production code, from campaign sites used by millions to AI tools for patent professionals.
- Think
- Philosophy taught me to ask why. Every interface I design starts with how people actually think.
- Design
- Briefs become user journeys, wireframes and prototypes in Figma before a line of code exists.
- Build
- Pixel-perfect, accessible builds: WCAG-compliant, keyboard-first and tested on every screen.
- Explain
- Now I give AI the same care, designing instruments that make complex systems legible.
07 · About
The person behind the instruments.
⚠ Error 429: Too Many Requests. Rate limit exceeded on WeWork lift mirrors. Retry-After: next floor.
Floor 8, About
Lift fit checks · drag them about
Fun factThe only thing I can order in sign language is coffee.
Don't hate meMy favourite desk set-up is laptop on crossed legs, back hunched.
Off the clockI spend a lot of my spare time writing and learning.
Hot takeI'd like to see AI have six cups of coffee and seventeen little walks a day.
SPECIMEN CARD● online
- Based
- Shoreditch, London
- Now
- Software Engineer, Clerq
- Origin
- BA Philosophy, King's College London
- Path
- philosophy → systems → full-stack → craft → intelligence
I started in philosophy at King's College London. From there I moved into systems, then full-stack engineering, and then spent years building games and campaigns that brands like PepsiCo, LEGO and Coca-Cola put in front of millions of people.
Now I'm at Clerq, building tools that let patent professionals see inside an AI search engine. Outside work, I build instruments that do the same for neural networks.
I like explaining hard things without jargon, and I do my best work around people: pairing, demoing and talking things through.
Experience
- Clerq● nowSoftware Engineer · AI patent intelligence
- Grand Central CreativeSoftware Engineer · campaigns and AI interfaces
- FreelanceFull-stack engineer · contract work for global brands
- Col8Software Engineer → Software Engineer II
- Two Hundred Rye LaneSystems administrator
- King's College LondonBA Philosophy
What it's like to work with me
“Seline is a true dream team player, with excellent communication skills that ensure everyone is on the same page.”
“Working with Seline brought me immense joy.”
“What really sets Seline apart is their time management and openness to feedback.”
08 · Contact
Let’s build something legible.
Hiring for AI products, data visualisation or full-stack work? I’d love to hear what you’re building.