berk.ai
← Work

MobTouch · Product

MyGut.io — AI health companion for IBD

AI-assisted health platform that helps people living with ulcerative colitis and other chronic digestive conditions capture daily reality, recognize patterns, and share a clearer picture with clinicians.

Role: Founder / Product engineer

  • AI for healthcare
  • Product design
  • Responsible AI

Origin

The story

MyGut.io began as a deeply personal project. Someone I care about lives with ulcerative colitis, and managing the condition meant trying to reconstruct an incomplete story from scattered pieces of information: meals, medications, symptoms, bowel movements, pain, fatigue, possible triggers, and changes that happened days or weeks apart.

I built MyGut.io as an intelligent health companion for people living with inflammatory bowel disease and other chronic digestive conditions. Its goal is to help patients and caregivers capture what is happening each day, recognize patterns over time, and communicate a clearer, more complete picture to their healthcare providers.

Instead of requiring users to manually record every detail, MyGut.io is designed to make tracking feel natural — photograph a meal, describe symptoms conversationally, log a bowel movement, record medications, or simply explain how they are feeling. AI helps organize that information into a structured, longitudinal health record.

The platform brings meals, symptoms, medications, bowel activity, pain locations, and daily observations together in one place, transforming them into timelines, calendars, trends, and insights that help reveal meaningful changes over time.

MyGut.io is not intended to diagnose or replace medical professionals. Its purpose is to help people become better observers of their own health and give clinicians richer, more organized information for making informed decisions.

The long-term vision is a personalized health companion that understands each individual’s baseline, detects meaningful changes, helps prepare for medical appointments, and reduces the burden of managing a complex chronic illness.

Context

The problem and my role

Problem

Managing IBD often means reconstructing an incomplete story from scattered pieces — meals, medications, symptoms, bowel movements, pain, fatigue, and possible triggers across days or weeks. Most trackers treat these as isolated data points instead of a connected longitudinal picture.

Role

Designed and built MyGut.io end-to-end as a full-stack, AI-assisted health-tracking platform: multimodal capture, structured clinical data models, timelines and trends, appointment prep, and responsible AI guardrails that never claim diagnosis.

Constraints

  • Must reduce data-entry friction for people who may be fatigued or in a flare
  • Preserve clinically meaningful structure without forcing medical-form UX
  • Separate health tracking and education from diagnosis
  • Handle sensitive personal health information with privacy, consent, and controlled sharing
  • Keep AI insights explainable and tied to original observations

System

Architecture

System map

From daily observations to a clearer clinical story

Natural input becomes structured, traceable health information that can be reviewed over time and prepared for a healthcare conversation.

  1. 01

    Patient input

    Conversation, meal photos, symptoms, medications, pain, and daily context

  2. 02

    AI conversation

    Understands natural language and asks context-aware follow-up questions

  3. 03

    Symptom engine

    Normalizes observations into a structured ontology while preserving the source

  4. 04

    Longitudinal health record

    Connects meals, bowel activity, symptoms, medications, and lifestyle factors

  5. 05

    Timeline and trend analysis

    Compares changes with the individual’s historical baseline

  6. 06

    Provider reports

    Produces explainable summaries for appointments and controlled sharing

Execution

What I built

  • Built a unified daily health timeline combining meals, bowel movements, symptoms, medications, pain, energy, hydration, sleep, and contextual notes.
  • Shipped AI-powered photo meal logging that turns meal images into structured nutritional and ingredient records.
  • Designed conversational symptom capture so people can describe how they feel naturally instead of navigating rigid medical forms.
  • Implemented comprehensive bowel movement tracking covering frequency, consistency, urgency, blood, pain, mucus, and related observations.
  • Added medication tracking for maintenance therapies, biologics, steroids, supplements, dosage changes, and adherence.
  • Created an interactive body map for location, severity, and progression of pain or discomfort.
  • Built calendar/timeline views plus trend analysis correlating meals, symptoms, medications, bowel activity, and lifestyle factors.
  • Designed personalized flare scoring based on deviations from an individual’s baseline, plus AI-generated appointment summaries and shareable reports.
  • Modeled a structured symptom ontology and an intelligent question engine that asks personalized follow-ups instead of fixed questionnaires.
  • Applied safety-oriented triage messaging and AI guardrails that clearly separate tracking from diagnosis.
  • Architected the stack with Next.js, React, TypeScript, PostgreSQL, Drizzle ORM, secure auth, image pipelines, and cloud object storage.

Tradeoffs

Technical decisions

Conversation in front, structure underneath

Decision
Let people describe experiences naturally while AI maps those observations into a normalized health model.
Why
Rigid forms create friction during illness, while structured records are still necessary for timelines, correlations, and useful summaries.
Tradeoff
Natural language is ambiguous, so interpretations must stay traceable to the original observation and invite correction.

Personal baseline over generic thresholds

Decision
Evaluate trends and flare signals relative to each person’s own history.
Why
Chronic conditions vary widely; meaningful change is often a deviation from an individual norm rather than a universal number.
Tradeoff
Personalization needs sufficient history and careful language so an observation is never presented as a diagnosis.

AI observations, not medical conclusions

Decision
Use deterministic guardrails and safety-oriented escalation around AI extraction, summaries, and insights.
Why
The product should help patients observe and communicate—not replace clinical judgment.
Tradeoff
Conservative boundaries limit how assertive the assistant can be, which is appropriate for sensitive health information.

Value

Product impact

User impact

Reduces the burden of recording a complex condition and helps patients and caregivers see a more complete story across days, weeks, and appointments.

Business impact

Demonstrates an AI-native healthcare product model where multimodal capture, structured data, explainability, privacy, and safety are designed together.

Results

Outcomes

  • Live product at mygut.io focused on IBD and chronic digestive disease management.
  • An AI-native healthcare experience where multimodal capture, structured records, and longitudinal insights are first-class — not a bolted-on chatbot.
  • A personal-to-product story that connects software architecture, AI, and healthcare engineering to a problem that became deeply personal.

Notes

Notes from the system

AI throughout the experience

Artificial intelligence is woven through meal recognition, symptom extraction, follow-up questioning, summarization, and trend detection — coordinated with deterministic application logic so the experience stays reliable and explainable.

Baseline-aware, not one-size-fits-all

Insights are grounded in each person’s history. Flare scoring and trends emphasize meaningful change from an individual’s normal — not generic population thresholds.

Reflection

Lessons learned

  • Building MyGut reinforced that AI should reduce friction rather than increase complexity.
  • In sensitive health workflows, preserving the link between a person’s original words and the system’s interpretation is as important as the insight itself.

Tools

Stack

GPT-5Multimodal AILangGraphTool callingStructured outputsNext.jsReactTypeScriptPostgreSQLDrizzle ORMCloud object storageSymptom ontologiesLongitudinal health recordsResponsible AI