Oliver Stevenson

Portfolio

Selected project work

Five standout builds from Oliver's GitHub projects, chosen for originality, shipped proof, technical depth, and product clarity.

Reviewed
55 repos
Featured
5 projects
Prizes
$5,199 USD
01

First place Superteam Imperial AI Agent Hackathon, $3,000

Relix

Live Source Prize
Relix AI Growth Employee live application screenshot

A founder-facing AI Growth Employee that connects GitHub, a website, X, analytics, and a wallet, then turns shipped product work into approved growth campaigns.

Stack Next.js, TypeScript, Anthropic, Solana devnet, GitHub OAuth, X OAuth
Core merit Specialist agents bid, deliver, and get selected inside the workflow
Prize proof 1st of 65 submissions for a Superteam agent economy bounty
On-chain proof Devnet settlement and agent-signed payout capabilities
02

Second place Opik hackathon, $850

Proompter

Source
Proompter prompt evaluation dashboard screenshot

A full-stack prompt evaluation and optimization platform for teams that need measurable prompt quality instead of guesswork.

Stack Next.js, Supabase, Python, Opik, LiteLLM
Core merit Multi-metric prompt scoring with experiment history
Product depth Team auth, A/B testing, chunked evaluation pipeline
Proof Hackathon placement and working product screenshots
03

Hackathon prize, $1,349 USD equivalent

Aletheia

Live API Frontend Backend
Aletheia social deduction game tavern scene

An AI-powered social deduction mini-game for the WeChat/Luffa runtime where generated statements create a replayable imposter hunt.

Stack JavaScript mini-game client, Express backend, OpenAI
Core merit Dynamic rounds instead of static trivia
UX polish Animated dialogue, accusation flow, audio feedback
Live proof Render backend endpoint used by the client
04

Built at a YC hackathon

Puffley

Source Demo
Puffley today screen with the day's craving count Puffley active Continuity Plan screen Puffley craving screen quoting the clinician's plan Puffley recovery progress and trends screen Puffley clinician-readable Recovery Passport screen Puffley consultation review and grounding screen

An iOS vaping-cessation app where a clinician configures a patient's Continuity Plan by talking; every line the app later says back to the patient is traced to a verbatim quote, checked twice, independently, so nothing ungrounded ever reaches the moment of a craving.

Stack SwiftUI, SwiftData, Claude Sonnet 5, ElevenLabs, Cloudflare Workers, Node.js
Core merit Dual-verified grounding: every plan item traced to a real quote, checked server- and on-device
Product depth Consultation capture, adaptive plan, craving intervention, judgement-free progress tracking
Proof 1,402 tests across ~38k lines of Swift, demo video filmed from a real UI test
05

Production-style mobile agent

Signal

Demo Live API Source
Signal demo thumbnail showing Opik evaluation traces

An iOS learning agent that decides when saved content deserves active recall and when it should protect the user's focus.

Stack Swift, Share Extension, TypeScript backend, OpenAI, Opik
Core merit Agentic decisions tied to real user attention
Evaluation Online traces, policy-aware judging, privacy controls
Proof Physical iPhone flow plus deployed backend