Not a CRM
No pipelines, no status stages, no pressure loops. The app avoids the feeling that relationships need to be managed like tasks.
Rehello
A gentle social memory app for introverts and people who find socializing tiring. Rehello helps them capture the fragments they remember, use GPT-5.6 to shape a messy note into a reviewable recall card, and feel more ready for the next conversation — without the pressure of a CRM.
Next.js 16 · TypeScript · GPT-5.6 · OpenAI Responses API · Zod · PWA
- Problem
- Most people tools are built for networking, sales, or pipeline management. That framing makes remembering people feel transactional, especially for users who already find socializing tiring.
- Built
- A mobile-first, local-first app with three capture paths: a zero-API demo, GPT-5.6 Quick Remember with a visitor-owned API key, and the original guided flow.
- Result
- A live, launch-stage product that makes its AI moment immediately explorable while keeping saved people, moments, and reminders in the current browser.
3Capture paths
7Structured fields
0Demo API calls
1Review before save
Context
Designing for people who want to remember others, not manage them.
Rehello started from a very different emotional premise than most networking tools. The problem was not "how do I track leads?" It was "how do I remember someone I just met without turning that person into a record?"
I wanted to explore a product that supports social memory for people who feel shy, overwhelmed, or mentally overloaded after meeting someone new. That meant the product needed to feel more like a quiet notebook than a productivity system.
The goal was to help users feel more human and more prepared in future interactions, not more optimized.
Product Framing
Shifting the category from contact management to social memory.
Most existing tools assume a transactional mindset: follow-up cadence, pipeline stages, or generic note fields. That felt wrong for this audience. Rehello needed a different product language, different interaction tone, and different priorities.
Not a social coach
The first version stays focused on memory support. It helps users remember what matters and gently prepare, without pretending to teach them how to perform socially.
Warm by design
The visual system, copy, and interactions all aim to reduce anxiety. Every surface is meant to feel calm, lightweight, and forgiving.
Local-first, with a clear AI boundary
Saved history stays in the browser. Only an explicit Quick Remember submission sends the current note and a visitor-owned API key through the server to OpenAI for that attempt.
Core Loop
Building one clear memory loop around real social situations.
The product is shaped around a simple sequence: capture what you remember, review what matters, and return to it before the next conversation. The newest version supports that loop without forcing every visitor into the same level of effort or API use.
-
01
Capture
Visitors can explore a fixed demo with no API call, turn their own messy note into a draft with GPT-5.6, or use the original nine-step guided flow one question at a time.
-
02
Review and save
Quick Remember returns seven structured fields, but nothing is saved automatically. The user reviews the card, adjusts anything that feels wrong, and then chooses whether to keep it locally.
-
03
Recall and reconnect
Person profiles, recall cards, Prep prompts, and in-app reminders bring useful context back at the right moment. No streaks, no scores, and no guilt-driven notifications.
Key Decisions
Design decisions that made the app feel more like a product.
Choice before commitment
A ready-made result proves the value without an account, key, or API charge. Users who want more control can choose Quick Remember or the original guided prompts.
Local-first, not local-only
People, encounters, reminders, Prep topics, and demo data stay in browser storage. The interface explicitly explains that a real Quick Remember request sends only the current note and one-time key through the server to OpenAI.
One bounded AI job
GPT-5.6 does one visible task: transform a 20–1,200 character note into a seven-field draft. Structured output constrains the shape, while the prompt tells the model not to invent facts or infer sensitive traits.
Human review before persistence
Generated drafts can still be incomplete or wrong. Rehello makes the result editable and requires a deliberate save, keeping the user — not the model — responsible for the personal record.
System Thinking
Creating consistency across browser storage, server boundaries, and mobile behavior.
Although the app is small, a lot of product work lived in the system details: how people sort, how reminders surface, how backup and restore behave, how demo data stays separate from paid API use, and how a one-time API key is cleared after an attempt.
I also treated the phone-sized shell as a shared system rather than a collection of individual pages. Stabilizing dynamic viewport height, safe-area behavior, horizontal containment, and bottom navigation made the experience more dependable across routes and narrow mobile widths.
The UX promise depends on more than warm visuals. The storage model, AI boundary, review step, and mobile shell all need to tell the same story: calm, explicit, and under the user's control.
What I Built
What the MVP includes today.
- Three capture paths: a ready-made zero-API demo, GPT-5.6 Quick Remember, and a nine-step guided flow
- A reviewable seven-field recall-card draft generated through the OpenAI Responses API and validated with a strict Zod schema
- Home, People, person profiles, Recall, Prep, in-app reminders, and Settings surfaces
- Browser-local records with manual JSON backup and restore, sample data, onboarding replay, and PWA install metadata
- A visitor-owned API key flow that does not persist the key, uses
store: false, and returns no-store responses - Vitest coverage plus CI gates for clean install, tests, lint, and production build
AI in Product & Process
How AI appears in the product and the build process
In the product, GPT-5.6 performs one bounded transformation: messy post-conversation note to structured recall-card draft. The server uses the OpenAI Responses API, low reasoning effort, strict structured output, and store: false. The generated card is always reviewed before it becomes a saved memory.
In the process, I used AI as a rapid product-building partner across naming, framing, front-end implementation, debugging, testing, and documentation. It accelerated execution, but it did not define the product point of view.
- AI helped with — generating code, exploring alternatives, debugging flow issues, and speeding up verification
- I was responsible for — product positioning, feature prioritization, privacy boundaries, interaction tone, and acceptance decisions
- The collaboration enabled — a fast path from concept to a working, documented product without giving up human judgment
Current Boundaries
Being explicit about what Rehello does not promise.
- There are no accounts, cloud database, or cross-device sync; saved records belong to one browser profile.
- Reminders appear only inside Rehello. There are no push, email, SMS, calendar, or operating-system notifications.
- The app has PWA install metadata, but no service worker or offline guarantee.
- Live Quick Remember is bring-your-own-key. API usage belongs to the visitor's OpenAI project, and generated drafts still require review.
- This is a launch-stage portfolio product, not a production-grade service; the public route does not yet have repository-level distributed rate limiting.
Reflection
What this project demonstrates.
Rehello shows how I think about product design beyond screens. The work spans category framing, interaction language, a visible but bounded AI role, privacy communication, responsive behavior, testing, and the judgment required to make a small system feel emotionally coherent.
The next meaningful step is still real-user testing: learning whether the capture-review-recall loop genuinely reduces anxiety before social re-encounters. The live MVP already demonstrates a strong point of view, but the most important outcome will be whether that point of view helps people in practice.