declassified
unlocking years of raw course feedback buried in Fizz's anonymous platform
Role
Product Designer
Timeline
3 Weeks | Feb 2025 - Mar 2025
Team
Solo
context
Fizz is Stanford's primary anonymous social media platform.
Fizz is an anonymous social media platform where college students can communicate with everyone at their school anonymously. At Stanford, though course evaluation isn't an intentional aspect of Fizz, students learn more from Fizz posts about classes than official course evaluations because people share things they'd never put in a university form in real-time: "Week 3 in this class was absolute chaos," "The professor is so forgiving," "Half the class is ready to drop."
The Problem: Years of brutally honest course insights were buried under endless scrolling.
user RESEARCH
Students already use Fizz as an unofficial course evaluator.
8
User interviews
3
Live observations
12-15
Posts opened per search
I observed 8 students searching for course reviews in real-time. The pattern was consistent: search for class code, scroll through 100+ posts, try to remember useful ones, search again for related course features (quarter, exams, attendance, etc), repeat.
Key Insight:
The search for candid course information on Fizz is a search for advice, not answers. Students rely on previous anonymous interactions to find the advice from their peers that they can usually only get by word-of-mouth.
➜
Students opened 12-15 posts before finding useful information.
➜
They wanted specific insights (workload, grading, prerequisites), but their searches returned everything.
➜
Most valuable posts had keywords like "grading," "time commitment," "skip if...".
➜
Students opened 12-15 posts before finding useful information.
➜
Recency didn't matter; students valued new and old posts equally as long as they contained relevant information.
THE GAP
Official evaluations miss what students actually need to know.
Official evaluations happen after the quarter ends. Fizz captures frustrations in real-time.
my SOLUTION
AI-powered categorization that preserves authentic student voice.
I created Declassified, an additional in-app course evaluation feature within Fizz that would provide an extensive yet concise one-stop-shop for the niche course information that students are already seeking out.
1
Smart Categorization
AI tags posts by topic: Workload, Grading, Teaching Style, Prerequisites
2
Relevance Ranking
Posts ranked by recency + engagement + keyword match
3
Smart Categorization
AI-generated summaries: "Students say: Heavy workload but worth it"
Declassified analyzes thousands of Fizz posts and surfaces relevant insights based on what students actually want to know without losing the raw, honest voice that makes Fizz valuable.
A few highlighted screens from Declassified user journey
content design
AI = navigation layer, not replacement for authentic voice
Key Tension:
Fizz's entire value comes from unfiltered anonymity. How do you organize years of candid posts without sanitizing the chaos that makes them trustworthy?
WRONG APPROACH
Hide posts behind sterile AI summaries (Loses the authentic student voice)
Ex. "This course has a heavy workload and strict grading policy."
STRONG APPROACH
AI surfaces most relevant posts first, summaries provide starting point (preserves authenticity while adding structure)
Ex. "Students say: Heavy workload but worth it" + 8 original posts below
The principle: Good design accelerates user goals without replacing the value they came for. Students get faster answers AND unfiltered peer experiences.
Key Features within Declassified
validation
User testing showed 80% faster task completion
80%
Faster search time (3 min → 30 sec)
73%
Fewer steps to find info
6
Usability tests conducted
407
Positive responses from current Fizz users
I conducted 6 usability tests with QuantUX heat mapping and cognitive load evaluation (foot-tapping accelerometer test). Results showed significant improvements in task completion time and user satisfaction (the task being the ability to find whatever the user considered to be "enough" information to decide how they felt about taking a course).
Additionally, I posted the mockup of my solution on Fizz to gauge excitement in current users; I received 404 upvotes in 12 hours and 3 comments, all of which were positive.
impact
Validated concept with executive buy-in
Though Declassified remains a concept, the response validated the design direction:
➜
Posted mockups on Fizz, users responded with enthusiasm
➜
Presented to Fizz CEO (Teddy Solomon) & Head of Product (David Vasquez)
➜
"We're surprised we hadn't thought of this before" — Executive feedback from presentation
➜
Testing demonstrated 80% improvement in course research efficiency
What I Learned: The biggest challenge wasn't the time constraint or prototyping AI integration; it was preserving trust in the content. Every design decision came back to: How do we add structure without sanitizing authenticity? How do we make AI helpful without making it the source of truth? How can we leverage AI's capabilities to boost what we get out of our face-to-face interactions rather than simply attempting to replicate them on Fizz? That tension between organization and chaos shaped every design decision I made.
Additional info