Research agent & tool for content
Turning content guesswork into tested theories
Planning videos meant hours of manual research and a lot of guessing. I designed and built a research tool in code that collects public video data over time. I use it as a dashboard and as an AI research agent I brainstorm with. The more data it collects, the better the ideas get.
Impact
videos collected and tracked over time
channels in the research
niches researched so far
Challenge
Every video started with manual research: scrolling, saving links, and trying to remember what I'd seen. It was slow and unvalidated. I had theories about why some videos took off, but no way to test them.
I needed a quick way to check my hunches against real data.
Designing for two ways of working
I'm the user, and I work two different ways depending on the task.
With an AI agent
For deep research and brainstorming. The agent runs the research and brings back evidence, so I can test ideas as I go.
On my own
For quick ideas. I browse the research in a dashboard, spot signals the AI might miss, and add ideas myself.
What I needed
Before building anything, I wrote down what my old process was missing.
Proof behind ideas
A way to check a theory against what actually happened.
Someone to think with
A partner to bounce ideas off and push back on weak ones.
Room for my instincts
Automation that supports my judgment.
Gets smarter over time
Research that builds on itself.
History is the value. The longer the tool collects, the more reliable its signals get, so it runs in the background.
Design process
I mapped the workflow before writing code, so every feature had a job. Then I built it in small pieces with AI-assisted development, fixing whatever slowed me down.
- 1. Collect
- Public video data, on a schedule
- Stored over time
- 2. Explore
- Browse niches and topics
- Spot what's gaining attention
- 3. Test a theory
- Ask a question, get evidence
- Compare against the history
- 4. Save and plan
- Ideas saved with their research
- Turned into video plans
Key decisions
Two front doors, one source of truth
The dashboard and the AI agent read and write the same research and idea bank, so nothing gets lost between them.
Evidence travels with every idea
Each saved idea keeps the research behind it, so I can see why it looked promising.
The AI suggests, I decide
Answers show the data behind them, and nothing gets added to my plans without my approval.
Built to run on a budget
Daily limits and spending guardrails are built into the system, so it can run in the background without surprises.
Key features
Three pieces that work together, from raw data to a video plan.
Research dashboard
Browse niches, channels, and topics, and spot standout videos at a glance.
AI research agent
Ask questions in plain language and get answers backed by the collected data.
Idea bank
Every idea is saved, sorted, and linked to its evidence, ready to plan.
Key outcomes
- Faster research
- Research that used to be manual now collects in the background, ready when I need it.
- Better ideas
- Ideas and writing get stronger the more data the tool collects.
- Tested theories
- I can check a theory about performance against real data, quickly.
Results so far
I use the tool regularly to research and plan videos for my channel, which launches soon.
Research collects in the background
Shared by the dashboard and the AI agent
Reflection
What I'd do next: give the dashboard more visual polish. What I'm watching: how researched ideas perform once my videos launch, so the tool can learn from my own results.
More of my work
- 1Read →
Charge-off payment plans
Simplifying payment plans for agents and customers
- 2Read →
Repaying after repossession
Helping customers understand and resolve what they still owe
- 3Read →
AI customer service assistant
Giving a collections chatbot a voice customers trust
- 4Read →
Design system & research templates
Giving the team shared building blocks instead of blank pages
- 5See all →
All work
Every project, plus additional work and experiments