

QuestGiver
A little more magic in the everyday. Point your iPhone at the world and turn ordinary objects into fantasy quests, relics and lore with local AI.
Forward deployed AI engineer
Eoin Tolster · AI engineer & product builderI build AI tools that solve real problems — from property document review and production systems to published iPhone apps and games.
Based in Melbourne / Made of curiosity
ET—03 / Orbital workshopTransmissions from the workshopWatch my builds on YouTube ↗
A little window into my world.
01 / Out of the workshop
A selection of things I’ve taken from an idea to something people can actually use or play.


A little more magic in the everyday. Point your iPhone at the world and turn ordinary objects into fantasy quests, relics and lore with local AI.

Property contracts, made clearer. At Lawlab, I helped build the AI review system across the interface, backend and AI workflow. Explore the product today.
Released on SteamFleet battles, a galactic campaign, and a pirate rebellion. A long-held ambition to ship a commercial game, made real.
Released on SteamCommand armies of flocking creatures, capture strategic nodes and use the terrain to your advantage in a single-player campaign.
Independent apps / iPhone
Useful tools, new ways to learn, and small worlds to get lost in.
Local AI
A local AI workspace for chat, document memory and research on your iPhone.
Learning
Interactive visual explanations that show why the maths works.
Memory
Piano patterns, language and science challenges for practising memory.
Hand tracking
An open palm to aim. A fist to shoot. Your hand becomes the controller.
Virtual pet
A little companion to feed, play with and care for in a cosy world.
Space exploration
Pilot a spacecraft, trade at stations and explore a procedural galaxy.
02 / Applied AI / Lawlab
Buying a property involves a lot of paperwork. I built an AI assistant that brings the important information together, checks its output and helps a person review it.
Approximate review times from my 2024 Lawlab work, as reported in my CV. The system supports a human reviewer’s decision.
The product todayExplore BuySecureImportant details were spread across lengthy property documents. Finding and reviewing them took time.
In a small engineering team, I built the interface, backend, AI workflow and deployment — connecting document reading, information extraction and checks into one review system.
Reviewers could get to the relevant information much sooner, with validation and human review built into the workflow.
OCR and PDF interpretation, structured extraction, document retrieval and model comparison, supported by a full-stack application and deployment. The work focused on reducing errors and giving reviewers information they could check.
Work through the documents, collect the relevant information and prepare it for review.
Each document is another place to look.
The assistant reads the documents and brings relevant details into a structured view for a person to check.
Illustration with fictional sample content, not a screenshot of the Lawlab product or live AI processing.
03 / The path here
From radio networks and field hardware to production AI. I work across the whole journey: understanding the problem, building the product, and making it work in the real world.
Download the full CVForward Deployed AI Engineer
Innovation Director
Senior Software Engineer · AI R&D
Applications Specialist
Software Engineer
Customer Support Engineer
04 / The curiosity department
Mixed reality, graphics, sound and playful software. A few of the things I’ve built to see what’s possible.

Virtual reality
A recorded look inside an experimental virtual reality museum.
Watch the preview
Mixed reality
Virtual objects, real surroundings: an early HoloLens experiment.
Watch the previewGraphics / JavaScript
A raycasting experiment: how simple rays turn into a navigable world.
Open the experimentSpace / Strategy
An early browser experiment in star systems, navigation and galactic strategy.
Open Galaxy ConquestPhysics / Play
An experiment in motion, rolling balls and building a playful challenge.
Open Ball RunNeural networks
Draw digits and explore a small neural network built in JavaScript.
Open the digit experimentSound / Interaction
A browser music experiment for exploring patterns and sound.
Open Music MakerFamily / Product
Fluqi / KidBank: an experiment in chores, pocket money and family routines.
Visit Fluqi / KidBankI share the technology I build, test and learn from — from AI experiments to worlds you can explore.
Visit my YouTube channel ↗
QUESTGIVER / BUILT FOR YOUR POCKETA little window into my workshop.
05 / Hello from Melbourne
I’m happiest moving between a difficult problem and a working prototype, then doing the less glamorous work that makes it useful to someone else.
My route here has taken me from Ireland to New Zealand to Australia, through radio systems, mapping, mixed reality and AI. Outside the build, it’s astronomy, sci-fi and strategy games. There’s usually another idea on the go.
Next transmission
Let’s talk about AI engineering opportunities, product collaborations, or creative projects.
Independent product / On-device AI



Screenshots from the published App Store listing.
Make an everyday object the beginning of a small adventure. QuestGiver interprets camera input and creates fictional quests, collectible relics and lore for a personal codex.
I built the iPhone app around local inference with llama.cpp, GGUF and Gemma. The challenge was bringing a camera-based AI experience into a product someone could carry and use.
A published iPhone app that connects my interests in practical AI, mobile software and imaginative worlds.
View QuestGiver ↗Lawlab / February — October 2024
Conveyancing review involved lengthy documents and a time-consuming manual workflow. The opportunity was to help reviewers get to the relevant information faster.
I built the review system across frontend, backend, AI orchestration and hosting in a small engineering team. The work combined OCR and PDF interpretation, document extraction, chunking, embeddings, retrieval and multiple model providers.
Validation, structured extraction, model comparison and human review were central to reducing hallucination risk. The system supported a reviewer’s judgment with information they could check.
Document review time reduced from approximately three days to around three minutes, as described in my September 2026 CV.
A high-level project account. The homepage illustration uses fictional sample content.
Read the case study ↓Published iPhone app / Local AI
A local AI workspace for chat, document memory and research on your iPhone.


Real screenshots from the published App Store listing.
View on the App Store ↗Published iPhone app / Learning
Interactive visual explanations that show why the maths works.


Real screenshots from the published App Store listing.
View on the App Store ↗Published iPhone app / Memory
Piano patterns, language and science challenges for practising memory.


Real screenshots from the published App Store listing.
View on the App Store ↗Published iPhone app / Hand tracking
An open palm to aim. A fist to shoot. Your hand becomes the controller.


Real screenshots from the published App Store listing.
View on the App Store ↗Published iPhone app / Virtual pet
A little companion to feed, play with and care for in a cosy world.


Real screenshots from the published App Store listing.
View on the App Store ↗Published iPhone app / Space exploration
Pilot a spacecraft, trade at stations and explore a procedural galaxy.


Real screenshots from the published App Store listing.
View on the App Store ↗