Product proposal

RapportArc

From first glance to lasting chemistry.

RapportArc is a specialized LLM-assisted training platform: learners log into a single-page app, follow guided courses with persistent progress, complete structured assessments and long-form exams with autosave, and receive rubric-based AI feedback. Interactive drills combine visuals and prompts; a persona simulator rehearses realistic threads. This page summarizes the product, architecture, and includes static interactive demos of key scenarios.

What RapportArc is

Digital product: subscription training courses delivered as an SPA—not a dating marketplace.
Progress & identity: enrollments and completion state per user account (backed by a database in production).
Assessments: standard forms (radio, checkbox, dropdown) plus long-form text with draft cache, resubmit, and feedback history.
LLM pipeline: qualitative submissions are scored and coached against versioned rubrics and scenario context.
Interactive modules: multimodal prompts (image, text, optional video) to elicit quick, creative responses—then optional AI commentary.
Girlfriend simulator: multi-turn chat with tunable personas (cold lead, app match, professional context, post-date texting, long-term rapport).
Dashboard: trends for scores and rubric dimensions over time.

Learner

Sees courses, completes modules, reviews feedback history and charts.

Platform

Auth, content delivery, progress, async jobs for LLM, storage for drafts.

AI layer

Rubric feedback, simulator turns, safety filters, prompt versioning.

Product design specification

The authoritative document covers ethics, user journeys, logical architecture, component list, data model, NFRs, and roadmap.

Open ProductDesignSpecification.md (full text)

Component map (summary)

Layer	Pieces
Client	SPA shell, course UI, forms, drills, charts, simulator transcript
API	Sessions, enrollments, submissions, simulator threads, analytics queries
Data	Relational DB, object storage for media, optional cache for drafts
Workers	Async LLM jobs, retries, cost caps, structured JSON output validation

System structure

Flow: browser SPA talks to an API over HTTPS; the API fronts a primary database, identity provider, and an async queue for LLM work. Media is served from object storage with signed URLs.

┌─────────────────────────────────────────────────────────────────┐
│                        Client SPA (browser)                      │
│  Auth · Course shell · Forms · Drills · Charts · Simulator UI    │
└────────────────────────────┬────────────────────────────────────┘
                             │ HTTPS / JSON
┌────────────────────────────▼────────────────────────────────────┐
│                      API (BFF / REST)                            │
│  Progress · Submissions · Simulator · Analytics                  │
└─────┬──────────────────┬────────────────────┬──────────────────┘
      │                  │                    │
      ▼                  ▼                    ▼
┌───────────┐    ┌───────────────┐    ┌──────────────────┐
│  Identity │    │  Primary DB   │    │  Object storage  │
│  (OIDC)   │    │  + drafts     │    │  (media)         │
└───────────┘    └───────┬───────┘    └──────────────────┘
                         │
                         ▼
                 ┌───────────────┐
                 │  Job queue    │
                 └───────┬───────┘
                         ▼
                 ┌───────────────┐
                 │ LLM pipeline  │
                 │ rubrics + sim │
                 └───────────────┘

Below are front-end-only previews with static or locally simulated data—no backend calls.

Scenario: sign in and land on home

RapportArc Train · Practice · Review

Password

Scenario: course list and module progress

Your courses

Messaging Arc 62%
Long-term Romance Not started
Foundations Complete

Messaging Arc — Module 3

First message from a cold slate

Lessons: Intro · Video · Drill · Quiz · Qualitative exam · Simulator checkpoint

Progress saved to your account.

Scenario: interactive drill (visual + creative association)

Training goal: rapid, playful ideation from a neutral image—used as a warm-up for conversation starters. (Fictional humor only; real-world respect always applies.)

Stock-style portrait

Prompt (playful): “In three quick lines, invent three absurd fictional ‘crimes’ she might have committed in a sitcom—no real judgment, just wit.”

Scenario: structured quiz

1. When texting after a first date, what is most aligned with polite, attractive follow-up?

Send a generic “hey” with no reference to the date. Reference something specific you both shared, plus a light next-step. Wait indefinitely with no message.

2. Pick signals that suggest rapport is building (choose all that apply).

She asks you questions back. She shares small details about her day. She only replies with single-word answers forever.

3. Scenario tone

Scenario: long-form exam + cached draft + LLM feedback

Autosave, submit, and resubmit; feedback is stored in history. Below is static mock feedback.

Autosave: idle

Scenario: progress dashboard (static chart data)

Rubric scores over time (demo)

Scenario: persona simulator (canned thread)

Select a scenario; messages advance from a static script to show the UX. Production uses LLM + persona system prompts.

Simulator Persona: Maya

Demo advances through predefined lines; your sends are echoed to the transcript.