GiANTbEAN — doTERRA Ecosystem AI Infrastructure

Muse — an AI companion that remembers you,
and keeps people at the heart of wellness.

A warm, Thai-native wellness companion grounded in a phytochemistry knowledge graph and built compliance-first — today in the hands of a small pilot community in Thailand.

Built by GiANTbEAN. What follows is what we've built, how we keep it honest, and where it's going — the graph Muse reasons over, the evaluation harness behind her, and the road toward an operating system for the PIPES journey.

16 / 19
Next.js 16 + React 19 — current-generation, not legacy tooling
≤3 hops
Knowledge-graph traversal, capped by architecture rule
360
Labeled samples in the golden evaluation set
0 / 30
PDPA-leak regression gate — hard-blocked in CI
MVP
Live pilot community in Thailand — real product, honest stage
Companion

Muse — a companion who remembers, not a chatbot

Muse is who Thai wellness seekers actually talk to inside doHEALER — mobile-first, warm, natural, Thai-native. Her job is to turn curiosity into a confident, informed decision, without pressure and without overreach.

She also remembers. Muse carries a three-tier memory — the current conversation, a set of persistent personal memories, and a knowledge-graph journey — so what matters to a person isn't forgotten between sessions. The longer someone walks with Muse, the better she understands them: a companion that grows with the user, by design.

That restraint in her voice is architectural, not a personality trait. Muse is scoped to structure-and-function language — how something feels, how it fits a routine — and is built to never frame a recommendation as diagnosis, treatment, or cure. That boundary is enforced downstream by the evaluation harness below, not left to a system prompt's good behavior.

a person This conversation what matters right now Personal memories persist between sessions Knowledge-graph journey her understanding of a person's walk — grows over time
Three tiers of memory around one person — the current conversation, persistent personal memories, and a knowledge-graph journey that grows over time.
Retrieval

Grounded in a graph, not a guess

Underneath Muse sits a Neo4j knowledge graph connecting oil constituents, phytochemistry, and body systems — paired with vector search (pgvector, over serverless Postgres) to find the right entry point for a question.

An answer is synthesized only from what the graph can actually reach. It can't wander three unrelated facts away from the question and call it an answer — the traversal itself is capped.

Question embed pgvector search nearest match Oil entry · hop 0 hop 1 Constit- uent hop 1 hop 2 Body system hop 2 hop 3 Context hop 3 · last allowed K-HOP CAP — ENFORCED BY RULE ? hop 4 — never traversed
A question is embedded, matched to an entry node by vector search, then traversed hop by hop through the graph. The 3-hop cap is an architecture rule, not a suggestion — the graph is structurally unable to reach a 4th-hop node.
Lead intelligence

FORMHD — what a great sponsor does, made teachable(Family · Occupation · Recreation/Relation · Money · Health · Dream)

The read a caring leader gets in a first conversation — the pain point, the readiness, the right kit to suggest — is what FORMHD gently captures from a real conversation with Muse.

And when a member steps up as a HEALER, Muse helps them create their own FORMHD reads and VIP Links — so growing a network starts from a genuine human conversation, not a cold list. Connection first; the structure follows.

a real conversation distilled FORMHD read a structured read for a caring leader
pain_pointsup to 5, ranked
desired_outcomesup to 3
budget_signalLOW · MID · HIGH
readiness_scorenumeric
recommended_kitSKU match
confidenceper-field
Evaluation

How we know Muse isn't making things up

We evaluate Muse the way a regulated product should be evaluated — not the way a demo is.

GOLDEN SET

360 labeled samples

6 extraction fields × 4 leader-rank tiers × 15 samples — the regression baseline every model change is checked against before it ships.

AGREEMENT

Cohen's Kappa, not vibes

Model output is scored for statistical agreement against human-labeled ground truth. A number, not a feeling, decides whether quality moved.

LEAK TEST

Adversarial PDPA suite — 0/30, every build

Thirty leak-attempt prompts fire on every change, checking Muse never surfaces internal source references or personal data. It's a CI gate that blocks the build, not a hope.

Model change Golden set 360 labeled samples Cohen's κ vs human labels every build PDPA LEAK GATE — 0 / 30 REQUIRED any leak — the build is blocked Ship
Every model change replays the golden set, is scored for statistical agreement against human labels, and must pass the adversarial PDPA suite — a CI gate that blocks the build on any leak — before it ships.
Compliance

Built for a regulated, global network

doTERRA operates across markets that each regulate wellness claims differently. This architecture already treats that as a first-class problem, not an afterthought.

Thailand policy source Malaysia policy source USA policy source Compliance source-of-truth modeled at the data layer, per market one Muse — the right rules for each market
Each market's policy is its own compliance source at the data layer — one companion, governed by the right rules wherever she speaks.

Compliance source-of-truth, per market

Thailand, Malaysia, and USA policy are modeled as separate compliance sources at the data layer — not one global prompt hoping to cover every jurisdiction.

Internal references never reach the user

Muse is architecturally blocked from surfacing internal-only source material to an end user — enforced by the same adversarial suite above, not a content filter bolted on after the fact.

Every admin action is audited

Nothing in the admin surface happens without a written audit trail — a requirement, not an option, at the routing layer.

Infrastructure

Real infrastructure, honestly staged

The stack underneath Muse was chosen for what it will need to carry — serious infrastructure under an MVP, ready to grow with the community it serves.

ApplicationNext.js 16 · React 19
Relational dataPostgreSQL (Neon, serverless)
Semantic searchpgvector
Knowledge graphNeo4j Aura
CacheUpstash Redis
AuthAuth.js
MonitoringSentry
DeploymentVercel
Live today

LianLian Pay Thailand is integrated end-to-end — RSA-signed, webhook-verified — and live for our pilot community.

live pilot

Payments were built in from day one — the MVP is real end-to-end.

Where this is going

An operating system for the PIPES journey

PIPES — Prepare · Invite · Present · Enroll · Support — is the rhythm doTERRA leaders already teach. We're not building a different path; we're building the AI that becomes a team's leverage power along it. Everything above is the foundation, and the road ahead extends Muse from one person's companion into the connective tissue of a team — always with consent, always person to person.

consent consent Wellness seeker today — walking with Muse Wellness Advocate Leader
One spine, extended along the journey — wellness seeker to Wellness Advocate to Leader — with explicit consent at every tier.

Oil Card — in the MVP now

Every doTERRA oil's story, guidance, and media gathered in one card — easy to find, easy to learn, easy to buy. Muse hands the right card into the conversation at the right moment.

Imprinting — designed

A first-meeting interview where Muse gets to know a new user — so she remembers them from day one, and every later conversation builds on who they actually are.

PIPES System OS — in architecture

The same knowledge-graph spine extended along the PIPES journey — from wellness seeker to Wellness Advocate to Leader — so an upline's Muse can support a downline's growth at every step, from Prepare through Support. The AI is the leverage power; the human is the connection.

Home oil shelf → Home Oil Club — shelf live, club envisioned

Members' own oils already inform Muse's suggestions today. The vision opens that shelf into a club view, so a leader can see what oils live in their community — and help them get used, not forgotten in a drawer.

Pure-to-Pure Exchange — designed

A sealed, new-only oil exchange inside a team, backed by a trust ledger — members trading what they have for what they need, keeping value circulating inside the community.

What we believe

We plant leadership — and let people grow.

GiANTbEAN believes leadership is planted, not appointed — and that real success is growth, from the inside out.

dōTERRA means “Gift of the Earth” — the phytochemistry the land offers through its plants, distilled into every drop of essential oil. That earth is what we love: a gift that spreads home to home, person to person. Our role is to keep that ground seeded — placing a seed of success in a person's hands every day, with Muse beside them as a companion who grows as they grow. Day by day, dōHEALER cultivates the habits of a dōTERRA HEALER.

And we believe this matters now more than ever. In the age of AI, the most precious things a network is built on are human relationship and trust. So we draw the line deliberately: let AI do the complex work — so people are free to do what only people can do. Build genuine bonds.

dōHEALER logo dōHEALER logo

www.doHEALER.com  ·  Private Alpha

Founder & Builder

Ratchawitth “Pao” Suwannajinda

doHEALER has one founder, and he builds it with his own hands — at the intersection where technology, leadership, and the inner life meet.

Outside the company, Pao serves as a volunteer dhamma speaker in Dhammanava “Wang” — a royal-initiative program under the Dhammanava Sikkhalai Foundation — bringing practical Buddhist wisdom to students, universities, civil servants, and women’s prisons across Thailand. More than 8,000 people have sat in those rooms. He teaches without charge, because he believes wisdom is planted the way leadership is: person to person.

Professionally, he has spent more than fifteen years designing business models and organizational ecosystems — including leadership-development and performance-culture systems inside the network-marketing industry, where he learned how leaders are actually grown. Today that experience is aimed at AI: doHEALER’s architecture — the knowledge graph, the evaluation harness, the multi-agent system behind Muse — is his own hands-on work.

Dhamma in the heart, systems in the hands. That combination is why doHEALER draws its line where it does: AI carries the complexity, so people are free to build trust.

Pao speaking on stage in the Dhammanava Wang program, with an audience of raised hands in the foreground.
Pao teaching in the Dhammanava “Wang” royal-initiative program — rooms of hundreds, across Thailand.
What we'd like to explore

We believe healing spreads person to person.

doHEALER is an MVP today, in the hands of a small pilot community — and we're honest about that. What we're certain of is the direction: technology that empowers people and helps build a new way of network marketing, one grounded in genuine human connection. We'd love to explore what that could look like together.

Two hands gently cradling dark soil with a young green seedling growing from it, over a forest floor.
GiANTbEAN — the team behind doHEALER (Aura Muse)