Sentiment-Aware Reflective Intelligence

MINDNOOK

The art of reflection,DECODED.

MindNook is a journaling platform built on a published five-layer NLP framework. It scores every entry against your own emotional baseline — not a generic average — models how your mood shifts over time, and knows the difference between a hard day and language that needs real support. Every AI reflection explains its reasoning, and every layer of analysis is yours to switch off.

Begin Writing Read the Paper

Personal Baseline Modeling

Crisis-Aware Safety Layer

Explainable AI Reflections

MindNook reflective journaling interface

Personal Baseline ModelingCrisis-Aware Safety LayerExplainable AI ReflectionsFive Independent Privacy ScopesPublished IEEE PreprintLLaMA 3.3 70B InferenceLongitudinal Trend Modeling35 Passing Tests Personal Baseline ModelingCrisis-Aware Safety LayerExplainable AI ReflectionsFive Independent Privacy ScopesPublished IEEE PreprintLLaMA 3.3 70B InferenceLongitudinal Trend Modeling35 Passing Tests

The Problem

Most journaling apps store your words.
None of them understand you.

Traditional journals are passive notebooks. They can hold a thousand entries and still tell you nothing about how your mood is actually trending, whether today genuinely differs from your baseline, or when a rough patch is starting to look like something more serious.

No Personal Reference Point

Your mood gets compared to nobody — or to a generic population average that has nothing to do with your actual history.

No Sense of When to Step In

Most AI companions either miss real distress entirely or overreact to a bad joke and an ordinary hard day.

No Reasoning You Can Check

You get a response, never the "why." There's no way to see — or challenge — how an AI reached its conclusion.

The Solution

Reflective writing, turned into
measurable self-understanding.

MindNook runs a published five-layer framework on every entry, wraps it in a crisis-aware safety layer, and shows its work at every step — so reflection becomes something you can actually track.

01 — Baseline

Compared Against You, Not a Population

Every entry is scored against your own historical mean, reported as a personal deviation — with a fixed population reference shown alongside it, never in place of it.

02 — Language

Pragmatic and Goal-Aware Analysis

Classifies what you're actually doing when you write — venting, asking for help, stating a fact — and checks it against the goals you've set for yourself.

03 — Trend

Longitudinal Trend Modeling

A recurrent model attends over your recent history and classifies your trajectory as stable, improving, declining, or cyclical.

04 — Safety

Crisis-Aware, Hyperbole-Literate Screening

A dedicated safety layer tells "this ruined my whole week" apart from language that signals genuine risk, before anything reaches the AI.

05 — Intelligence

Nook AI, Reasoning Transparently

Every reply is built from your live framework values and moderated before you see it — and can be expanded into its layer-by-layer reasoning on request.

06 — Control

Calibrated to You, Not a Default

Set your own tolerance for false alarms versus missed signals during onboarding, and let the system adjust — within bounds you control — from your own feedback.

MindNook dashboard and reflective canvas

Inside MindNook

One entry. Five layers of understanding.

Every time you write, MindNook simultaneously scores sentiment, classifies intent, updates your longitudinal trend model, checks alignment with your stated goals, and calculates the right response strategy — surfaced on a dashboard built to show change over time, not just a list of past entries.

Personal and population baseline, side by side
See your own deviation next to a fixed reference point — never one instead of the other.
Mood timelines and tone breakdowns, visualized
A Chart.js dashboard renders your emotional trend, vocabulary growth, and pragmatic mix across your entire history.
One-click reasoning on every AI response
Expand any reflection into the layer values that produced it, and flag the ones that don't feel right.

How It Works

Four steps to reflective intelligence.

Write Naturally

Open a distraction-free reflective canvas and write. No prompts, no forced structure, no interruptions.

Five Layers Run Instantly

Sentiment, pragmatics, temporal trend, goal alignment, and response strategy all compute the moment you save.

Safety Checks First

A hyperbole-aware, crisis-screening pass runs client-side before the AI ever responds — with no network dependency.

Reflect With Nook AI

Talk to a companion whose tone adapts to your real state, and that can show its reasoning whenever you ask.

Published Research

Architecture

Built on a published
sentiment-aware framework.

MindNook is the reference implementation of a five-layer framework for sentiment-aware reflective writing systems, published as an IEEE TechRxiv preprint. The platform combines client-side NLP, schema-constrained LLM inference, and serverless edge architecture into a psychologically grounded journaling experience — with graceful degradation to local computation whenever a model call fails.

Read the Paper

MindNook research and reflective writing

Framework Layers

Tests Passing (Vitest)

LLaMA 3.3

70B via Groq Inference

Edge

Supabase Functions

Privacy

Private by design.

MindNook prioritizes psychological safety. Every inference type is independently controllable, and you keep full ownership over your reflective history.

Protected Inference Pipelines

Every AI call runs server-side through authenticated Supabase Edge Functions. Your entries are never processed by a third-party client SDK directly.

Five Independent Privacy Scopes

Sentiment, pragmatic, temporal, goal-inference, and full-history chat access can each be switched off — and turning one off stops that computation entirely.

Full Export and Deletion Rights

Export a complete JSON copy of your entries and analysis at any time. Deletion is confirmation-gated and cascades across every linked table.

FAQ

Straight answers, before you start.

What makes MindNook different from a regular journaling app?

MindNook scores every entry against your own historical baseline instead of a generic average, then runs four more layers on top — pragmatics, temporal trend, goal alignment, and response strategy — that shape how the AI responds, not just what it says.

Is this based on real research?

Yes. MindNook is the reference implementation of a five-layer framework published as an IEEE TechRxiv preprint, and the app computes all five layers on every entry rather than approximating them.

What happens if I write something concerning?

A client-side, hyperbole-aware safety layer checks for genuine risk language before anything reaches the AI. Figurative language like "this ruined my week" is treated differently from language indicating real risk, and confirmed risk surfaces a crisis-resource banner.

Can I turn off specific types of analysis?

Yes. Sentiment, pragmatic, temporal, goal-inference, and full-history chat access are five independently toggleable privacy scopes, and disabling one stops that computation rather than just hiding its output.

What AI model powers Nook AI?

Nook AI runs on LLaMA 3.3 70B via Groq, called through Supabase Edge Functions, with every reply passed through a moderation check before it reaches you.

Can I delete my data?

Yes, at any time. A confirmation-gated deletion cascades across every user-linked table, and a self-serve export gives you a full JSON copy of your entries and analysis first.

Understand your mind,
not just your memories.

Every entry becomes a data point in a framework that's published, tested, and built to explain itself. Start writing — or read the research first.

Start Writing Read the Paper