The problem
There's a lot of film out there, and it's hard to know what to watch next. Options are scattered across streaming services and cinemas, and recommendations usually lean on what you've already watched, so they tend to be predictable.
The challenges:
- What you feel like watching depends on your mood and the moment
- Availability is scattered across services and cinemas
- Recommendations based on past viewing are predictable
- Film buffs often want to know where a film sits in film history, not just what's similar
The solution
To address these issues, I designed an app to help curate movie watchlists and find films to watch.
Key features
- Curate a watchlist with watched state, synced across devices
- Filter by mood on six scales
- Explore a film's sources and influences
- Find similar films by theme, genre or style
- Find more films by the same key collaborators
Benefits
- Track the films you intend to watch and mark them off as you watch them
- Pick something to suit the moment, by mood rather than just genre
- Learn where a film sits in film history through its sources and influences
- Find new films to watch starting from one you already like
- Follow filmmakers you enjoy through their other work
- See where to watch a film
Learning about users & the problem space
I interviewed 5 people of different backgrounds and levels of interest in film.
Levels of engagement with film:
Film buffs
These are people that use specific criteria/sources for recommendations. They follow specific filmmakers and trusted experts, curate watchlists, see film as a way to teach culture, and analyse more deeply.
Casual watchers
These are people for whom trends/popularity & social media drives the selection of films. They judge films on the basis of genre and analyse the positives, rather than dwell on the negative aspects.
Other observations:
Mood & mindset
The time of day, environment & current feelings influence choice of film. In addition, subjective expectations of a film can influence the reception of that film.
Availability
The availability of films on streaming services or in physical cinemas drive a lot of film watching decisions.
Social dynamics
People may watch films they otherwise wouldn't when with friends. They may even enjoy bad movies ironically in such a situation. Furthermore, most like to discuss the films afterwards.
Understanding the existing market
I examined 5 different products, including streaming services and movie recommendation apps:
Characteristics of competitors
- Tinder-style interfaces: common but inefficient for new users
- Personalisation: recommendations based on past choices
- Categorisation: often excessive and not particularly meaningful
- Demographics: apps vary in level of information provided about films, providing either too much or too little information for the different cohorts
- Social curation features: often tailored for specific social groups like couples or friends
- Streaming availability: most apps indicate where films can be streamed
Who are the users and what are their goals?
Based on my research, I was able to define two archetypal personas, each with challenges, frustrations and goals.
These are their goals:
Hoss, the film buff
- Wants to more easily curate a list of interesting films to watch and select an appropriate film from it to watch at any given opportunity
- Wants to deepen and broaden his knowledge of film and film history
- Wants to broaden his film watching horizons, as feels like he is missing out on potential gems
- Wants to find more films by filmmakers whose work he has enjoyed
- Wants to share his film suggestions and knowledge with his immediate friends
Amy, the social movie watcher
- Wants to find new, popular/trending movies or TV series that would provide a good social watching experience with her friends
- Wants to find interesting new movies and TV series that are popular/trending so she can be part of the conversation
- Wants to understand what her friends are interested in watching together, so she can organise mutually satisfying group watches with them
- Wants to bond with her friends over group watching experiences
How do we address the opportunities?
I identified some key challenges to address based on my research.
From research I knew that The Movie Database (TMDB) offered APIs that provide film data required to implement many basic features. I took that into consideration when prioritising features for MVP.
For the key challenge:
For the key challenges I brainstormed some solutions:
...I brainstormed the following solutions:
Helping users to curate and track watchlists
Solution:
- Save films to watchlist and mark as watched/unwatched
Helping users to find interesting and satisfying films to watch that they haven't seen before
Solution:
- Basic info about each film
- Suggestions based on previously curated films
- Explore films by cast/crew/genres
After MVP:
- Contextual info about films, such as history & film movements
- Explore films by movement/history
Helping users to find films to watch that fulfil the audience's current psychological wants and needs
Solution:
- Filter movies by genre/mood/experience
- Suggestions by genre/mood/experience
- View film availability
After MVP:
- Watchlist sharing & collaboration
- Automatic suggestions based on time/mood
Designing & testing iteratively
I started with a sitemap, user flows, sketches and low-fidelity wireframes, and tested them with users and other designers.
I then created a high-fidelity interactive prototype covering three key flows. I tested these flows with 5 potential users and iterated on what I learned.
Findings:
- New users needed a first-time experience
- Some terminology was confusing
- The relationships between elements on the search screen were unclear
- Users wanted watched films kept on the watchlist as a history
- More information about each film was important to make decisions, including runtime and critical ratings
The solution was rated relatively high for ease of use
5 – 7 out of 7
Users felt it would make a good replacement for the fragmented interfaces of streaming services
However, it could be further enhanced:
- Sharing & group curation
- More information about each film
- Lift trending/popular categories and filters
Design progression
Crafting the product identity
For this project I created a brand & design system, which I iterated concurrently while designing the product. Feedback from other designers informed the development.
Identity development
Rethinking the concept as an AI-driven web app
The original concept wasn't tightly scoped and didn't stand out from existing products.
Along the way I came to realise that film history makes for much richer suggestions than taste alone. Knowing what a film grew out of, and what it went on to influence, leads to films that “people who liked this also liked” rarely would.
So I rebuilt it around that idea. It's still a work in progress, and for MVP I focused on the needs and goals of the Hoss persona.
Hoss wanted to…
Hoss wanted to…
...so I designed:
Curate a list and pick one to suit the moment
...so I designed:
- Watchlist with watched state, synced across devices.
- Mood filters on six scales, such as slow burn to fast paced.
Deepen his knowledge of film history
...so I designed:
- For any given film, list that film's sources and influences.
Broaden what he watches
...so I designed:
- For any given film, list similar films based on themes, genres, or style.
Find more films by filmmakers he likes
...so I designed:
- For any given film, list films by the same key collaborators.
I deliberately decided to leave out social features for MVP, as they would rely on network effects, which are not practical for a new product.
Tackling the design challenges
Shaping the AI suggestions
Suggestions start from a film, not a taste profile, to avoid predictable AI recommendations. AI covers themes, lineage and mood, and TMDB supplies metadata. It took four versions to balance strictness and reliability.
Benchmarking the AI results
I benchmarked documented film pairs that the usual recommendation services don't tend to surface:
- Blow Out was directly inspired by Blow Up and also shares ground with The Conversation. It checks that the app separates a clear influence from a looser connection.
- She Freak, an obscure 1967 film, has an established link to the far more famous Freaks. It checks that the app finds the connection for a film with very little written about it.
Designing the mood axes
I replaced fixed mood tags with six scales from -2 to +2, such as slow burn to fast paced. Each scale had to vary within a genre and not overlap with genre or another scale.
Balancing cost, speed and API limits
Two caches cut cost and delay: AI suggestions are kept until the prompt changes, and film details for 30 days to respect TMDB's terms. Lineage only loads when its tab is opened.
Building the app
I used AI iteratively throughout the design and implementation:
- Brainstorming: generated ideas and explored concepts, using my initial research as a starting point.
- Design: explored user journeys and layout options with it, such as the mobile filters.
- Development: co-created implementation plans with it, and executed them, surgically making manual adjustments.
- Testing and review: benchmarked the AI results and manually reviewed the implementation at each stage.
Tech stack:
Next.js, React, TypeScript · AWS Amplify, DynamoDB, Cognito, Systems Manager Parameter Store, CloudWatch · TMDB, OpenAI
