‹ back to portfolio

Social feed redesign

Fritter

Make the feed your own.

Three working features that give readers control over curation, browsing time, and the rewards behind their feed.

Research to implementation · Vue + TypeScript
Fritter feed with posts locked until the reader earns points Fritter chat screen where messaging a friend earns browsing time Fritter daily limit warning offering five more minutes or exit
Role
Solo Designer & Developer
Type
Research → Shipped Redesign
Timeline
7 weeks · Sep–Oct 2022
Context
MIT 6.1040 Software Studio

Overview

Fritter is a redesign of the social feed built around a single idea: people should be able to say what they want to see, and understand why they're seeing it. I ran the project end to end — teardown of the incumbent, two user interviews, six concepts, three shipped features, peer critique, and a working frontend and backend.

Twitter decides what you see and never explains how. How might we hand that decision back to the person scrolling — without making the feed worse?
User Research Concept Design Interaction Design Design Critique Vue + TypeScript MongoDB
Stage 01 · Understand

Tearing down the incumbent

Before proposing anything, I audited how Twitter actually behaves — what it rewards, what it hides, and where its design creates harm. I looked at accessibility affordances, engagement mechanics, and the way controversy propagates through quote-tweeting.

The finding that shaped everything after: the platform treats all engagement as positive signal. A quote-tweet written to condemn a post boosts that post exactly as much as one written to endorse it. The system has no concept of the user's intent — so outrage and agreement are indistinguishable inputs to the same ranking function.

What the platform gets right is worth naming too. Hashtags, bios, and threads make it genuinely possible to find people like you, and accessibility affordances have improved steadily. Any redesign had to preserve that discovery capability while fixing the ranking opacity underneath it.

Twitter's alt text editor for describing images
Accessibility affordances added over time — alt text authoring for images
A Twitter thread discussing identity and belonging across cultures
What works: threads let people find others navigating the same questions
Stage 02 · Plan

Defining who to talk to, and why

I wrote selection criteria before recruiting anyone, so the sample would stretch across the dimensions most likely to change how someone experiences a feed:

  • Personal background — lived experience that shapes what community means online
  • Academic background — inside and outside my own institution and discipline, so the sample wasn't a mirror
  • Belief background — people who identify strongly with a religious or political community, to test whether the feed builds or breaks echo chambers

I then deliberately recruited two participants who sat at opposite ends of usage: one light user who mostly posts, one heavy user who mostly browses. If both reported the same frustration, that frustration was structural rather than personal.

Stage 03 · Listen

Two interviews, opposite ends of the spectrum

Participant A · light user
The advocate who posts

Joined recently on a mentor's recommendation. Uses the platform to find community and share resources, and is deliberate about what she reveals. Values hashtags and bios as the connective tissue that makes finding people possible.

"If you want to talk to another disabled person, you can just search them up… through these hashtags you can find community and share resources."

Participant B · heavy user
The browser who lurks

Years of daily use, mostly scrolling for news, sport, and politics. Likes the speed and the ease of skipping past things. Aware the habit has gotten away from him, and frustrated that the feed gives him no dial to turn.

"I wish there's more user choice for what's popping up… it would be nice if user has input to a curated content, not just the algorithm."

Stage 04 · Synthesize

Where two very different users converged

Participant A wanted reach without exposure — community access without surrendering privacy. Participant B wanted speed without capture — quick information without the compulsive pull. On the surface, opposite problems.

Both, however, described the same underlying cause: they had no say in how the feed was assembled, and no visibility into why any given post appeared. Participant A put the mechanism plainly — the algorithm "does not tell you whether things are bad engagement or good engagement." Participant B described the consequence: "if you lean one way directly, you are probably going to interact with only people sharing the same mindset."

The insight that set the direction

A light sharer and a heavy browser, with nothing else in common, both asked for the same thing: explicit input into curation. Opacity wasn't a preference problem, it was a structural one — which meant the fix belonged in the mechanism, not in the settings menu.

Stage 05 · Diverge

Six concepts

Each concept got a stated purpose and an operating principle — what it's for, and exactly what happens when someone uses it. Three adapted existing patterns, three invented new ones.

Sketch of periodic user survey concept
Concept 01 · adjust
Periodic user survey

Ask people directly what they want more of, and whether they want opposing views — then curate from the answer.

Shipped
Sketch of screen time limit concept
Concept 02 · adopt
Screen time limit

A self-set daily cap, borrowed from iOS Screen Time, with a soft "five more minutes" override.

Shipped
Sketch of accumulating points reward concept
Concept 04 · adopt
Accumulating points

Earn browsing minutes by messaging friends, so the currency of the app is conversation rather than consumption.

Shipped
Sketch of disappearing content concept
Concept 03 · invent
Disappearing content

Fifty posts a day, chosen by topic, each vanishing as you scroll past it. Empty feed until tomorrow.

Cut
Sketch of public search history concept
Concept 05 · invent
Public search history

Exceed your limit and the app publishes your recent searches and most-visited profiles to your page.

Cut
Sketch of spinning wheel randomization concept
Concept 06 · invent
Spinning wheel

Spin to pick one of five randomized bundles drawn from the hashtags you follow.

Cut
Stage 06 · Decide

Three in, three out

The three I cut are the more interesting half of the decision.

ConceptCallReasoning
Periodic survey Shipped The only concept that gave direct, explicit input into curation — the exact thing both participants asked for. Daily framing also let intent change day to day.
Reward system Shipped Turned time on the app into something you earn through conversation, addressing the compulsion without simply locking people out.
Time limit Shipped Kept opt-in by design. It exists for people who want a brake, and stays invisible to those who don't.
Public search history Cut It works by shame, and shame is unevenly distributed — plenty of users simply wouldn't care, which defeats the mechanism for exactly the people it targets. It also cut against the goal of making the app feel socially safe.
Disappearing content Cut Ephemerality undercuts the reason to post at all: what's the point of being here if nothing you share is preserved? Participant A's use case — sharing resources others can find later — dies under this model.
Spinning wheel Cut Randomization makes the mechanism visible without making it controllable. It's transparency theatre, and the survey already delivered real agency.

Wireframing the survivors

With three concepts left, I moved into Figma to work out the actual flows — how the survey opens the day, where earned minutes surface while browsing, and what the limit does when it fires. Drawing the screens is also what exposed the conflicts that peer critique would later confirm.

Stage 07 · Specify

Specifying the system

Each feature was decomposed into concepts with an explicit purpose, state, and set of actions, then synchronized. Writing the state model surfaced conflicts between features long before any of it was built.

Overall data model showing concepts and their relationships
The data model — eight concepts and the synchronizations between them
Feature specification for user curated content
Feature 1 — User Frittered Content: auth, post, and survey concepts in sync
Feature specification for the reward system
Feature 2 — Reward System: friend, chat, and reward concepts in sync
Feature specification for the time limit
Feature 3 — Limited Countdown: time limit synced to browsing, not chat
Stage 08 · Critique

What critique changed

I reviewed four classmates' designs against concept independence, efficacy, simplicity, and synchronization, and had mine reviewed on the same axes. Four criticisms landed, and each produced a specific change:

CriticismTwo friends could spam each other with junk messages to farm browsing minutes, gaming the reward system.
→
ChangeAdded a dictionary check on outgoing messages, so repeated or meaningless strings don't earn points.
CriticismThe time limit could cut someone off mid-conversation — the app punishing the exact behaviour it rewards.
→
ChangeAdded a "Chat Only" mode. The limit now applies to browsing the public feed, never to messages.
CriticismThe Friend concept existed only to make rewards work, and its relationship to following was unclear.
→
ChangeFriends' posts now always appear regardless of survey answers, comments earn points too, and following became explicitly one-way.
CriticismFilling out a survey every single day is a high cost of operation for a small benefit.
→
ChangeYesterday's answers carry forward with one tap, and hashtags became mandatory on posts so the pool stays rich.

One I pushed back on

Two reviewers argued that public-by-default posting would induce anxiety and discourage people from posting. I kept it. The goal was never for everyone to post — quieter users can browse and use chat, both first-class paths through the app. Making posts public is what creates the shared surface that community forms on, and Participant A's entire use case depended on strangers being able to find her.

Stage 09 · Build

Shipping it

I built the design out as a working application — Vue and TypeScript on the front end, Express and MongoDB behind it.

User Frittered Content

The daily survey, and the feed it produces. You can always trace a post back to an answer you gave.

Fritter Reward System

Messaging a friend earns browsing minutes. Posts stay locked until you've spent some of them.

Limited Fritter Countdown

A self-set daily cap on browsing, with chat deliberately exempt from the clock.

Stage 10 · Reflect

What I'd do differently

I'd validate the reward mechanic with users before specifying it. I designed it from a principle — conversation over consumption — and only found the spam loophole once it reached critique. A handful of concept tests would have surfaced that a week earlier, and cheaply.

I'd also treat the daily survey as a hypothesis rather than a settled feature. Cost of operation was the criticism I most underestimated: I was so focused on giving people control that I didn't count what exercising that control would cost them every single morning. Agency isn't free, and a design that demands it daily has to earn the interruption.

Research participants are referred to as Participant A and Participant B. Names and identifying details have been removed or generalized; quotes are preserved except where they contained identifying information.