Clarify once
Ask one question only when the input contains multiple possible intents.
02 / Project
Course-recording retrieval tool
CuePoint
Turns a vague memory of a technical lesson into a verified, timecoded lecture moment.
- Context
- SCAD IXDS 720 · Spring 2026
- My Role
- Solo Product Designer
- Outcome
- Tested functional MVP
01 / Problem
A small answer can hide inside hours of video.
A 20-second answer can be buried inside hours of recorded instruction.
Students in software-heavy courses often remember that an instructor demonstrated a step, but not the lecture, wording, or timestamp. Recovering that one moment can mean scanning multiple recordings and interrupting the assignment already in progress.
The information exists. The problem is the cost of retrieving it.
02 / Insight
People remember tasks, not database keywords.
A recording archive is organized by weeks, titles, full videos, transcripts, and timestamps. A student remembers a question: “Where did the instructor explain how to update nested state?” CuePoint starts from that natural memory and translates it into a retrievable intent.
A timestamp is not enough. Users need evidence before they act.
03 / Scope
One course, one task, a controlled technical boundary.
In scope
Natural questions
Search using the task or problem the student can remember.
In scope
Verified moments
Show why a result matches before sending the user into the recording.
Out of scope
Production search
No live transcript pipeline, production backend, or live API.
04 / Solution
Search, verify, then return to the original lesson.
Ask a question
→
Clarify if broad
→
Review Best Match
→
Open Moment Detail
Best Match makes the recommendation legible before the user watches.
05 / Key decisions
Trust is designed between retrieval and action.
Go direct
Skip clarification when enough context already exists.
Show no-match
Refuse to force a plausible-looking answer outside the curated course data.
Moment Detail is the confirmation layer between a result and the original recording.
06 / System
The interface follows the data relationships.
The final model connects six entities. The segment is the key bridge: it turns a long recording into a retrievable instructional moment while preserving its course and video context.
01Course
02Recording
03Segment
04SearchQuery
05SearchResult
06SavedResult
07 / Testing
Two prototype stages answered different questions.
Figma alpha
Tested whether the search structure, result hierarchy, and core navigation made sense.
Functional MVP
Tested conditional clarification, more judgeable results, richer Moment Detail evidence, clear system feedback, and no-match handling.
08 / Outcome
A functional product arc with an honest implementation boundary.
I owned problem framing, scope definition, system logic, information architecture, interaction design, the data model, prototyping, testing, iteration, and final product communication.
The MVP proves the end-to-end interaction from natural question to verified lecture moment. Live ingestion, real-time transcripts, and production matching remain future technical work.