01
Turn reports into action, not just reading
The first value of AI is translating lab results into steps users can understand and start immediately.
Turn gut testing
into a growth-ready health service.
A Boilingwater AI solution sample for a gut-health management brand. We designed a custom AI system around report interpretation, personalized intervention, dietitian collaboration, and repeat purchase / retesting, so the model becomes part of the service workflow instead of just another chatbot.
6AI engines
Profile · report · plan · execution · consultation · growth
12week delivery
From problem alignment to pilot launch
4week programs
Default intervention cycle with room to extend
7×24basic Q&A
AI handles repeated questions while dietitians focus on higher-value care
ai-health.console · Report interpretationv3.2
Ms. Lin · microbiome report
Sampled 2026-05-02 · lab · parsed
Microbial diversity42% · low
Butyrate producers28% · insufficient
Gas-producing bacteria73% · high
AI interpretationSOP-NUT-014
Start with one action: raise daily fiber from 12g to 22g over the next two weeks, mainly from oats, kiwi, and psyllium husk. Do not introduce probiotics yet because high gas-producing bacteria plus bloating may worsen discomfort.
Fig. 01 · Dietitian console / AI interpretation output (desensitized)
Solution Brief01 / 09
The problem is not
sending users another report.
For gut testing, functional food, nutrition, and chronic-care businesses, delivering a report is only the starting point. The real business question is whether users understand it, take action, keep following the plan, and come back for retesting or repeat purchase.
02
Base recommendations on a real profile
Diet, exercise, sleep, allergies, and test history become one user profile, so recommendations fit the person.
03
Make plans professional and executable
Weekly and monthly guidance is decomposed into daily tasks, with adherence data feeding plan adjustments.
04
Free dietitians from repetitive work
AI handles interpretation, summaries, and common questions so human experts can focus on high-value care.
05
Make AI recommendations governable
AI must be auditable, observable, and constrained by business rules, not just capable of answering.
Diagnosis02 / 09
In health workflows, AI fails when
it looks useful but cannot be operated.
Before designing the solution, we worked with the client to identify the real bottlenecks and decide where AI should enter the workflow. These were the five highest-priority problems.
01
Communication Gap
The report is professional, but users still do not act
Lab reports are dense and technical. For the platform, sending a report is not the same as delivering value. If users cannot understand it, they will not execute or repurchase.
Related AI engine§ 03 · Engine B
02
Data Fragmentation
Lots of user data, but no decision-ready profile
Symptoms, diet, sleep, exercise, mood, allergies, and historical tests are scattered across forms, chats, and third-party systems. Without one profile, advice becomes generic.
Related AI engine§ 03 · Engine A
03
Execution Friction
A professional plan often fails because it is hard to execute
Conversion and retention depend on whether the plan respects preferences, cost, time, and difficulty, and whether it can adapt when a user drops off.
Related AI engine§ 03 · Engines C · D
04
Human Bottleneck
Dietitians are valuable, but repetitive work consumes them
Most high-frequency questions are repeated interpretation, common Q&A, and reminders. AI should absorb that work so experts can focus on risk judgment and deep service.
Related AI engine§ 03 · Engine E
05
Governance Risk
Without governance, smarter AI becomes a bigger risk
Health advice requires clear boundaries, evidence, traceability, and human takeover. AI must be audited, monitored, and constrained by business rules.
Related AI engine§ 04 · Governance
AI Solution03 / 09
Six AI engines mapped to six real business problems.
Each one ties to a verifiable business outcome.
This is not a universal chatbot. The solution is split into six engines with clear responsibilities, connected by one profile and execution-data layer to form an auditable service loop.
Engine A
Profile Engine
Turn scattered signals into a living health profile
Conversational questionnaires collect symptoms, habits, goals, and constraints, then combine diet preferences, allergies, routines, test history, and feedback into a continuously updated profile.
Delivered assets
- Dynamic health questionnaire
- Diet / exercise / sleep / mood / allergy data model
- Profile score, risks, and goal APIs
Engine B
Report Interpretation
Translate technical reports into actions users will follow
The engine structures microbiome diversity, abundance, and function indicators, explains them in plain language, links them to symptoms, and prioritizes next actions.
Delivered assets
- Lab report parsing from PDF / JSON / lab feeds
- Term translation, symptom linkage, and risk tiers
- Prioritized do / do-not-do action list
Engine C
Plan Generation
Move from generic advice to explainable personal plans
Based on reports, profile, and goals, the engine generates 4-12 week plans across diet, exercise, supplements, and mood, with evidence attached to every recommendation.
Delivered assets
- Plan templates across four dimensions
- Traceable evidence from indicator to rule
- Versioning, dietitian review, and one-click release
Engine D
Execution Engine
Break long interventions into daily executable tasks
Weekly and monthly guidance becomes daily task cards, reminders adapt to user preference and past completion, and adherence data feeds future plan adjustments.
Delivered assets
- Daily tasks for diet / exercise / supplement / mood
- Adherence tracking and interruption reasons
- Execution data back into plan tuning
Engine E
Consultation Engine
Let dietitians focus on high-value service
AI handles repeated interpretation, pre-consultation, and common questions. Out-of-bound or risky cases route to humans with a concise context summary.
Delivered assets
- FAQ knowledge base with user context
- Risk rules, human handoff, and pre-consult summaries
- Dietitian console for review and adjustments
Engine F
Growth Engine
Connect plan execution to retesting and repurchase
The engine recommends functional foods or supplements based on plan stage and goals, explains the reason, triggers retesting, and unifies service, commerce, and consultation metrics.
Delivered assets
- Stage-based product recommendations with reasons
- Retest and repurchase triggers
- Unified growth dashboard
Architecture04 / 09
We did not just connect a model.
We embedded model capability into a five-layer system.
Health workflows require more than general AI capability. The system must be explainable, traceable, and constrained by business rules. This is the engineering structure used in the project.
系统分层 · System Layers
L1
Data intake and profile layer
/ User contextLab reports, questionnaires, app behavior, and third-party health data are normalized into one profile store shared by downstream modules.
profile_storelab_reportsevents_stream
↓
L2
Inference and orchestration layer
/ AI engineSix engines run as independent services orchestrated by a shared workflow layer. Every LLM call uses structured input, retrieval context, and rule constraints.
orchestratorretrievalprompt_kitsrule_engine
↓
L3
Knowledge and rules layer
/ Professional boundariesNutrition knowledge, client SOPs, compliance boundaries, and risk triggers are versioned and auditable. AI responses must stay within approved knowledge.
knowledge_basepolicy_rulesrisk_triggers
↓
L4
Governance and observability layer
/ AI governanceEvery output is logged, scored, sampled, and traceable. Low-confidence or sensitive cases automatically route to human review.
audit_logeval_runshuman_handoff
↓
L5
Business and operations layer
/ Growth and serviceAdherence, commerce conversion, and retesting events flow back into future model decisions so recommendations and reminders improve over time.
growth_signalscampaign_engineretention_loops
Engineering Principles
01
Structured input and output
LLMs are not allowed to improvise freely. Inputs and outputs are constrained by clear schemas, which is the baseline for auditability in health workflows.
02
Retrieval plus rule constraints
AI needs access to the client knowledge base, but it must also be limited by whitelists, risk rules, and human takeover points.
03
Every inference is traceable
Inputs, context, model versions, outputs, confidence, and human intervention are stored for operational review and compliance inspection.
04
Execution data closes the loop
Adherence, drop-off reasons, and product feedback are not only BI metrics; they become inputs for the next round of plan generation.
Sample Outputs05 / 09
Three real work samples,
showing concrete AI output in concrete scenarios.
All samples are adapted from desensitized solution drafts. They show how AI works across report translation, plan generation, and dietitian collaboration.
Scenario
Ms. Lin receives a gut microbiome report but only sees a wall of technical indicators.
The report shows low microbial diversity, insufficient butyrate producers, and high gas-producing bacteria. The old workflow simply pushed charts to the user. They could not understand them and did not act.
- USER
- F · 32 · sedentary / high stress
- INPUT
- lab_report.json + 7-question intake
- GOAL
- Improve constipation / bloating
AI · output preview
stream · structured · auditedEngine B · user-facing interpretation
AI compresses 26 indicators into three next actions, each linked to symptoms and evidence.
- FACT
Your microbial diversity is around 2.1, in the low range. This means the gut ecosystem is relatively fragile and sensitive to diet changes.
- ACTION
For the next two weeks, focus on one action: raise daily fiber from about 12g to 22g, mainly from oats, kiwi, and psyllium husk.
- RULE
Do not start probiotics yet. Because your gas-producing bacteria are high and you report bloating, some probiotics may worsen discomfort before diet stabilizes.
- NOTE
Basis: report §3.2, questionnaire Q5, and SOP-NUT-014. This conclusion can be reviewed in My Plan.
schema · v3.2latency · 1.4s · model · in-house
Engagement06 / 09
A 12-week engagement path,
with a business-usable asset at every step.
We do not deliver a handover document plus a chatbot. Each phase produces assets the business can use directly, so AI enters the workflow step by step before launch day.
01
W01 - W02
Business problem alignment
- Interview business owners and dietitians to identify bottlenecks
- Map user journeys, dietitian workflow, and available data
- Define the five high-priority AI problems for this phase
Deliverable
Problem Map · a shared working document, not a slide deck
02
W02 - W04
Solution blueprint and prototype
- Draw the system map and data flow around six engines
- Define output schemas, knowledge structure, risk rules, and human takeover points
- Review prototypes with dietitians to keep AI inside boundaries
Deliverable
Solution Blueprint v1 · schemas and boundaries included
03
W04 - W08
Core engine implementation
- Build MVP versions of profile, report interpretation, plan generation, and execution engines
- Connect lab reports and existing client systems to run end-to-end samples
- Create an eval set to measure rewrite and correction rates
Deliverable
Demo-ready end-to-end sample with real-data replay
04
W08 - W10
Dietitian collaboration and AI governance
- Build dietitian console for summaries, review, revision, and handoff
- Launch AI governance panel for logs, sampling, version trace, and risk triggers
- Run three internal test rounds across five user journeys
Deliverable
Console v1 · dietitian and admin perspectives
05
W10 - W12
Launch, review, and growth pilot
- Gray launch and observe adherence, handoff rate, commerce conversion, and retest triggers
- Feed execution data into the growth engine and run two pilot campaigns
- Deliver next-stage roadmap with KPIs and milestones
Deliverable
Launch Report + Roadmap v2
Business Outcomes07 / 09
We do not measure this by
how smart the AI looks. We measure what each role gets.
The directional metrics below come from business goals defined with the client. Exact values vary by baseline, user volume, and operation strategy.
For users
User
They no longer receive a report they cannot act on.
- Get a clear improvement path
- Receive advice based on their own profile
- Complete long-term intervention through tasks, reminders, and feedback
For dietitians
Dietitian
Less repetitive work, more professional judgment.
- AI handles interpretation, Q&A, and summaries
- Risk points and priorities are easier to locate
- Human time goes to plan adjustment and relationship building
For the platform
Platform
A one-time test becomes a long-term operating asset.
- Improve post-report retention and plan adoption
- Improve adherence and service continuity
- Increase dietitian capacity
- Create more retest and repurchase opportunities
Directional Metrics
Plan adoption
More users start an intervention after reading the interpretation
Adherence
Daily task completion and continuation improve steadily
Repeated Q&A time
Dietitians spend less time on basic explanations
Retest / repurchase
More conversion opportunities after the intervention cycle
The real value of AI here is not answering one more question.
It is giving users, dietitians, and the platform a shared decision base that keeps improving.
— Boilingwater · Project Review
Applicable Scenarios08 / 09
This is not only for gut health.
It applies to every test-interpret-plan-execute-repeat loop.
This page presents one specific cooperation case. The structure, including six engines, five system layers, and a five-phase path, can be adapted to other health businesses by replacing reports, rules, and product systems.
Direct fit
Gut microbiome testing and intervention
The scenario shown in this case: intake, report interpretation, microbiome-related diet and supplement guidance, and retesting triggers.
Discuss this scenario
High fit
Nutrition and chronic-care management
For glucose, lipids, weight, and women's health workflows where AI turns indicator changes into executable lifestyle actions.
Discuss this scenario
High fit
Functional food / probiotic membership
Upgrade one-time purchases into assessment, recommendation, reminder, feedback, and subscription workflows.
Discuss this scenario
High fit
Medical checkup report interpretation
Extend annual checkups into continuous health-management touchpoints for clinics, online healthcare, and corporate wellness.
Discuss this scenario
Strong fit
Private-domain health advisor teams
Give community health teams an AI collaboration console to expand service coverage and standardize quality.
Discuss this scenario
Adaptable
Other test-intervention-retest businesses
Any workflow with assessment, interpretation, plan, execution, repeat purchase, or retesting can reuse the same structure.
Discuss this scenario
Why Custom Development
Why this type of project fits custom development
The value of health-management AI is not a generic chat box. It comes from embedding AI into workflow, data structure, service boundaries, and growth strategy.
What custom development really means
Adapting lab report structures, business rules, dietitian workflows, product systems, repurchase strategy, private-domain operations, risk control, and audit needs.
Multi-role collaboration without feature bloat
The usual roles are user app, dietitian console, and admin backend. The point is not more portals; every role must support a real AI-enabled business action.
Boilingwater does not sell “model integration.”
We deliver AI products that can actually support business growth.
Let's TalkIf you have a concrete workflow
If you have a concrete workflow
AI has not solved yet, let's evaluate the right approach.
The first strategy call is free. We will unpack the workflow, judge whether AI is worth using, identify the right technical route, and provide a practical initial plan and estimate within five business days.
Business email
contact@boilingwater.cnOffice
10F, South Tower, Kingkey Yujing Times, Longgang District, Shenzhen