Advise, don’t decide
The product never picks for you. It turns a messy dilemma into options, risks, and one low-regret next step — so the user still owns the choice, just with less rumination.
Project case
Live
Decision Assistant
A shipped AI product for everyday dilemmas: pour out the mess, get a low-regret next step, then review it later — so decisions stop vanishing into chat history.
Problem
The hard part is not more information — it is convergence.
I often get stuck on small-but-real choices: buy or wait, keep or quit, rest or push. ChatGPT can answer once, then the thread disappears. What I needed was a process that records the dilemma, forces a low-regret action, and makes later review possible.
Product
Not a smarter chatbot — a decision loop.
Decision Assistant wraps a warm “little-lamb” shell around a fixed loop: speak the dilemma, get a structured draft, take a low-regret action, then review. Over time, those records can become an observable decision profile.
- Pour out the dilemma in one breath
- AI drafts options, risks, and a low-regret action
- Save the record and review it later
Key calls
Three product judgments that shaped the build.
Process over one-shot answers
The gap versus ChatGPT is not a stronger model. It is a durable loop: draft → act → review → profile. Without review, it is just another suggestion box.
Pour-out first, forms later
V1 asked for title, category, options, fears, mood. That made an already anxious person fill a spreadsheet. The default became a single freeform dump; structure is inferred afterward, and the first screen leads with the conclusion.
Scope
What I owned end to end.
From positioning and prompt design to Web/Android delivery: product judgment first, then shipping with AI-assisted engineering.
Product / PRDLLM prompts & workflowWeb appAndroid buildDeploy & iteration
Outcome
A real demo, and a clearer bar for what is worth building.
The product is live on Web and Android. More importantly, the build reset my bar: in an AI-agent era, code gets cheaper, judgment gets more expensive. The next filter is habit — will I open this instead of ChatGPT when I am actually stuck?
Next
Want something similar shipped?
I help turn fuzzy ideas into usable AI tools: workflow design, prototypes, LLM integration, and lightweight product delivery.