← All Work

07 — AI Projects

AI Projects.

Tools I designed and built with AI — from marketing software to NLP analysis. Not prompts in a chat window: real, runnable apps where marketing and data thinking meet shipped product.

Project 01 · Built with Claude Code

AI Marketing Suite

A self-serve marketing tool that turns a single product brief — description, features, audience, screenshots — into a complete go-to-market kit. I designed the product, wrote the prompts and the scoring rubrics, and built the app: Python (Flask) + the Claude API, with structured AI outputs so the copy comes back clean and editable every time. Two working modules so far.

Module 01 Python · Claude API

Brand Studio — instant brand collateral

Enter a product once; get a full on-brand collateral suite back — each piece editable in-app and exportable to Markdown.

  • Brand foundation: positioning, value props, messaging pillars, taglines, tone of voice
  • A slide-by-slide pitch deck with speaker notes
  • A one-page brochure and launch-day social posts
  • Reads uploaded product screenshots to ground the copy
Module 02 Python · Claude API

Social Planner — calendar + predicted fit

From the same product brief, it plans a launch content calendar tuned to the audience — and predicts how each post lands with each persona.

  • AI-generated, fully editable audience personas
  • A multi-week calendar across LinkedIn, Instagram, and X
  • A predicted-fit score per persona, per post — clearly flagged as an AI estimate, not measured engagement
  • Approve & schedule posts, then export the calendar to CSV
Why it matters: this is the marketing workflow I know inside-out — brand foundation, collateral, audience targeting, content planning — rebuilt as a tool. It shows the strategy thinking and the ability to ship it as software with AI. The predicted-fit score is deliberately labelled an estimate of messaging fit, not validated performance data — honest by design.

Project 02 · NLP & Data Science

App Review Intelligence

A voice-of-customer tool that turns thousands of raw app-store reviews into a ranked list of what to fix and what to amplify. It pulls live reviews for any app, classifies each one by sentiment with a pre-trained transformer model, then mines the recurring themes behind the praise and the complaints — the analytical flip-side of the community and social work I've done for years.

Module 01 Python · Hugging Face

Sentiment Engine — feedback at scale

Pulls hundreds of recent reviews on demand and scores every one, turning a wall of text into a clear positive / neutral / negative picture.

  • Live review collection for any app via the store API
  • Per-review sentiment via a pre-trained RoBERTa model — no training data required
  • Sentiment cross-checked against star ratings as a built-in sanity test
  • Confidence score attached to each prediction
Module 02 Python · scikit-learn

Theme Miner — the “why” behind the rating

A label isn't an insight. This surfaces the actual phrases users repeat, so a team knows exactly what to fix first and what to lean into.

  • TF-IDF phrase extraction across positive vs. negative reviews
  • Top complaint themes and top loved-features, ranked
  • Charts plus a complaint word cloud for fast read-out
  • Exports a prioritised product-improvement list
Why it matters: this is the listening half of marketing made rigorous — the same instinct behind managing a 500K+ community, but built to scale to thousands of voices and output a decision, not just a feeling. Using a pre-trained model is a deliberate engineering choice: sentiment is a solved problem, so the effort goes into the insight, not reinventing the model.

What this demonstrates

Marketing strategy, shipped.

The interesting part isn't the code — it's the judgment behind it: encoding marketing thinking into a product, and knowing which signal in a pile of customer feedback actually deserves a team's attention. Strategy on one side, data on the other, both shipped as working software.

Product design Prompt engineering Brand strategy Persona & audience modelling NLP & sentiment analysis Hugging Face transformers Data visualization Claude API / structured outputs Python · Flask · scikit-learn Shipping with AI
See it live: both projects run as working code — happy to give a walkthrough or screen-share on request. The marketing suite — Brand Studio, Social Planner, and a live Review Radar that reads any app's current tone on demand — was built and iterated with Claude Code; App Review Intelligence also runs as the original notebook you can re-run on any app.
← Data Analytics Back to all work →