Portfolio
5 projects spanning 0→1 launches, growth strategy, viral design, and metric analysis. Each one shows how I think through a problem end-to-end.
0→1 AI Product Build · careerlantern.henryelue.space
Problem: Job seekers lack personalised, affordable tools for ATS-optimised resumes, realistic interview prep, and career transition planning. Generic templates and expensive coaching leave most users underserved.
What I built: Designed and shipped a full-stack AI career platform end-to-end: product strategy, architecture decisions, and feature delivery. 17 authenticated pages, 12 AI endpoints, and a complete auth system (JWT + Google/LinkedIn OAuth, bcrypt, rate limiting, CSRF protection).
AI Architecture: Initially deployed Llama 3.2 3B (self-hosted via Ollama, exposed through a Cloudflare Tunnel at a custom OpenAI-compatible endpoint), eliminating per-call API costs and maintaining full model control. Migrating to Gemini 2.5 Flash for improved output quality and reliability at scale. ElevenLabs scribe_v1 handles speech-to-text for voice interview answers (60s audio → transcribed text via MediaRecorder). A global system prompt enforces precision, ATS optimisation, career-level calibration, and strict JSON output discipline across all AI features.
Key technical decisions: FastAPI + Uvicorn (async Python for AI workloads) → Next.js 15 App Router (SSR + TypeScript); PostgreSQL + Alembic migrations; DigitalOcean Spaces with private ACL + presigned URLs for resume file storage; APScheduler for daily follow-up reminder emails; Mixpanel + PostHog for dual product analytics and session replay.
Product pillars shipped: AI resume builder (CV import → LLM parse → structured JSON → pre-fill; 3 bullet rewrite styles; ATS/JD match score 0–100 with keyword gap analysis) · Voice mock interview coach (streaming AI feedback, sample improved answers, answer library for 8+ scored responses) · Career intelligence (skills gap analysis, alignment score, transferable skills map for career switchers) · Cover letter generator · LinkedIn headline + About optimizer · Job application tracker with status pipeline and follow-up reminders.
0→1 Product Launch · Emerj LLC (noramum.app)
Problem: Expecting and new mothers face anxiety, isolation, and information overload with no consistent, personalised support system. Most AI tools also aren't built with the safety constraints a health-adjacent product requires.
What I did: Led 0→1 product development of a mobile-first AI companion, defining product vision, conducting customer discovery, designing core features (AI chat, milestone tracking, personalised daily check-ins, memory timeline), and owning execution from concept to App Store & Play Store launch.
AI Architecture: Built on OpenAI as the primary model with Gemini as a backup fallback, ensuring reliability and continuity if the primary model is unavailable. Backend powered by FastAPI; mobile app built in React Native for a single codebase across iOS and Android.
AI Safety & Guardrails: Designed and enforced explicit safety constraints in the system prompt. Nora is hard-blocked from giving medical advice. All AI responses include contextual reminders that information should be discussed with a qualified healthcare provider. This was a deliberate product decision to manage liability, maintain user trust, and keep the product in a supportive companion role rather than a diagnostic one.
Outcome: Shipped a live B2C product used by expecting and new mothers. Improved onboarding completion by 20% through optimised flows and experimentation; increased user engagement by 15% by prioritising AI chat and milestone-based personalisation.
Product Concept & Design · Figma Prototype
Problem: People want to connect around shared interests but existing platforms require sharing phone numbers or social media handles, which creates friction and privacy concerns that discourage genuine community formation.
What I did: Conceived and designed GroupInn from the ground up, defining the product vision, target users (clubs, cohorts, learning teams, associations, online communities), and core feature set. Designed four key pillars: Group Discovery (find communities by interest), Group Chat (real-time messaging), Recognition (highlight milestones and contributions), and Privacy Control (connect without exposing personal details). Built out the full concept in Figma, from user flows to high-fidelity screens.
Insight: People stay engaged with products when they find their people. GroupInn focuses on intentional, human connection, keeping interactions easy, safe, and organised without the complexity of mainstream social platforms.
Growth Strategy · Dec 2025
Problem: Audiomack's engagement was centred on library size rather than community, which limited organic growth and long-term retention.
Solution: Designed "Crew Charts", small listening groups where each Crew generates its own live music chart from member activity. The viral loop: listen, get prompted to invite friends, create or join a Crew, share the Crew Chart, new users onboard directly into a Crew. Shifted the product's emphasis from passive listening to community-driven music taste.
Success metrics defined: % new users from Crew invites, active Crews per MAU, weekly listening hours per user, Crew Chart share frequency, and 4-week retention of Crew users vs non-Crew users.
Root Cause Analysis · YouTube
Problem: A 30% drop in Average View Duration (AVD), a core engagement metric tied directly to revenue and creator monetization, needed to be diagnosed and resolved.
Approach: Applied a structured RCA framework: clarified scope with 7 targeted questions (onset, category breakdown, algorithm changes, ad load, Shorts vs long-form, regional differences), documented assumptions, then analysed data across user feedback, video analytics (CTR, retention), user behaviour, content trends, and competitor activity.
Root causes identified: Content quality decline, suboptimal algorithm recommendations, increased ad load, user shift to short-form (Shorts/TikTok), platform usability issues, and evolving viewer expectations.
Recommendations: Short-term: retune recommendation algorithm, A/B test ad placement, fix usability. Mid-term: lean into Shorts, incentivise quality creators. Long-term: exclusive content, interactive features (polls, Q&A, live streams). Defined AVD recovery rate and engagement rate as success metrics with iterative rollouts.
Get in touch
Open to PM roles, freelance product work, and conversations about interesting problems.
Connect on LinkedIn