HELP.Try HELP →
Case Study · 2025

HELP.
The closure engine
for loud minds.

A research-grade thinking workspace for people whose minds run loud. Where other chatbots end with an answer, HELP ends with a graph of what you know, what you're missing, and what you left open.

0
Titan reasoning layers
0
Gemini models routed
0+
Signature features
0%
RLS coverage
01 · The problem

Loud minds don't have an information problem.

They have a closure problem. The tab count keeps growing, the answers keep arriving, and nothing ever feels done.

Symptom 01
Over-research, under-decide
40 tabs open, no decision made. The research becomes the procrastination.
Symptom 02
Threads die on Sundays
The idea you loved on Tuesday is invisible by the weekend. There is no ritual to reopen it.
Symptom 03
Fluency mistaken for depth
Standard chatbots reward smooth answers. Loud minds smell it and disengage.
Symptom 04
No memory of shape
Nothing remembers how you think — only what you last typed.
02 · The insight

Five contrarian bets.

Each one, when we stopped fighting it, unlocked a piece of the product.

01
Loud minds want closure, not answers.
So every reply must name what's still missing.
02
Gaps are more valuable than citations.
The "Missing Node" is more prized than the "Sources" list.
03
Personas are permission slips to think differently.
Voices bind to threads so users can switch brains, not settings.
04
Linear chat is a bug for deep thinkers.
The Knowledge Graph is a first-class surface, not a menu item.
05
The moat is the shape of the user, not the model.
pgvector concepts + edges accumulate into a graph nobody can copy.
03 · The flow

Every message becomes a closed loop.

Titan reasons. The gap detector names what's missing. Expand Knowledge fills it. The graph remembers. The Return brings you back.

Stage 01
Think
Titan runs 6 layers, dissents with itself, and settles.
Stage 02
Connect
Gap detector names 3 non-obvious unknowns.
Stage 03
Shape
Expand Knowledge fills the gaps with real research.
Stage 04
Refine
Dots + edges land in your graph and pull you back.
04 · The product

Twelve features. One thesis.

Click any feature. Each one exists to close a loop that loud minds usually leave open.

6-layer reasoning

Titan Engine

What
Intuition → Antithesis → Synthesis → Expansion → Refinement → Polish.
Why it matters
Turns a chatbot into a thinking partner that disagrees with itself.
Feature 01 / 12
05 · Architecture

How it's wired.

Client + server functions + a swarm of Gemini models + a Postgres graph. Every table has RLS from day one.

Cognitive shape · data flow
Row-Level Security
On every public table. No policies = locked by default.
Signed image paths
Server-side ownership check before any vision call.
Weighted credits
message 1 · expand 8 · visualize 5 · steelman 6. Auto-refund on failure.
06 · The journey

From chatbot to closure engine.

Seven pivots. Each one killed a comfortable assumption.

Phase 01
Chat
Fluid dark UI, sliding mindmap panel, deep-research prompt parser.
Phase 02
Wisdom layer
thread_memory + agents; the bot starts to remember you.
Phase 03
Titan Engine
6-layer reasoning ladder; internal traces hidden from the reply.
Phase 04
Knowledge Graph
pgvector concepts + edges; cross-thread semantic recall.
Phase 05
Gap-first pivot
Show the Dots → Missing Node; Expand Knowledge stores knowledge, not links.
Phase 06
Constellation map
Starfield · focus mode · ghost gaps — the graph feels alive.
Phase 07
Market-ready
Credits, legal, auth, usage limits, teaser, landing.
07 · Decisions

What we killed.

Every good product is defined by what it refused to be.

KILLED
3-point template replies
Loud minds felt talked-down-to.
KILLED
Sources by default
Closure > citations. Sources only in Research mode.
KILLED
Titan progress in the UI
Kept internal. No '6 layers · 50.7s' leaking into the reply.
KILLED
Marketing landing page
The first click is a New Thought, not a signup form.
KILLED
Empty states on the map
Replaced with ghost gaps — the map always shows what's missing.
KILLED
Manual thread naming
Auto-rename from the first user message.
The thesis
The moat isn't the model.
It's the graph of you.

Every message you send sharpens a picture nobody else can copy — how you think, what you keep circling, what you refuse to close.

Close a loop — free← Home