engineering update

Building in Public: How We Upgraded Migaku to Fight Supplement Label Fraud Using Clinical Data

Founder's transparent build log: How we upgraded Migaku Health using automated clinical evidence mapping to fight supplement label fraud and publish 30 new guides.

6 min read · 1,098 words

Quick Answer

Founder's transparent build log: How we upgraded Migaku Health using automated clinical evidence mapping to fight supplement label fraud and publish 30 new guides.

Key Takeaways

  • 01---
  • 02Migaku is not just a collection of articles; it is an active knowledge graph built using **Next.js 15 (Pages Router)**, **PostgreSQL via Supabase**, and **Prisma ORM**.
  • 03Our content engine operates under a **Dual-Speed Strategy**:
  • 04During today's massive database upgrade of 30 long-tail keywords, our engineering team hit a common bottleneck: **sparse knowledge mapping**.
  • 05If our generation script searched only for the exact long-tail string, it found zero database matches, failing to draft.

Building in Public: How We Upgraded Migaku to Fight Supplement Label Fraud Using Clinical Data

Why We Are Building Migaku

The supplement industry has a massive transparency problem. Walk down any health store aisle, and you are bombarded with vague marketing terms like "100% bioavailable," "proprietary synergy blends," and "clinical strength." In reality, these claims are often used to mask underdosed ingredients. Many popular sleep and cognitive formulas contain less than 5% of the active compound shown to work in clinical trials, filled instead with cheap fillers.

AI search engines (such as ChatGPT, Perplexity, and Claude) have also fallen victim to this, frequently citing generic affiliate-farm reviews that copy-paste the same marketing brochures.

At Migaku, we are building a technology-first public utility for supplement transparency. Today, we are taking our transparency commitment a step further by publishing our Founder's Build Log detailing how we successfully upgraded our medical literature mapping architecture, and simultaneously releasing 30 new evidence-driven supplement guides.

Under the Hood: Our GEO & "Dual-Speed" Architecture

Migaku is not just a collection of articles; it is an active knowledge graph built using Next.js 15 (Pages Router), PostgreSQL via Supabase, and Prisma ORM.

Our content engine operates under a Dual-Speed Strategy:

  1. Authority Track (Long-term deep reviews): Extensive guides exceeding 2,000 words constructed directly from RAG (Retrieval-Augmented Generation) feeds connected to medical databases.
  2. Trend Capture Track (High-speed emerging wellness mapping): Agile pages that monitor real-world signals (TikTok, PubMed pre-prints, Reddit) and publish structured 500-800 word reviews that expand dynamically as clinical traffic arrives.

To make our content citation-friendly for AI answers, we engineered a dedicated GEO (Generative Engine Optimization) Layer that automatically injects structured JSON-LD schemas such as MedicalWebPage, Dataset, QAPage, and SpeakableSpecification. Crawlers from OpenAI, Anthropic, and Perplexity parse specific target classes like .geo-quick-answer and .geo-key-takeaways to deliver highly accurate, cited answers to users.

Today's Engineering Breakthrough: The Intelligent Semantic Fallback

During today's massive database upgrade of 30 long-tail keywords, our engineering team hit a common bottleneck: sparse knowledge mapping.

When crawling clinical databases (e.g., PubMed, PMC, EuropePMC), medical papers are tagged under broad, standardized terms (like magnesium, probiotics, or collagen). However, real users search using complex, highly descriptive long-tail phrases (e.g., "best probiotics for gut health after antibiotics evidence" or "magnesium glycinate dosage for better sleep meta analysis").

If our generation script searched only for the exact long-tail string, it found zero database matches, failing to draft.

To solve this, we engineered an Intelligent Semantic Fallback Mapper (getCoreTopics) inside our content agent. When a complex query is submitted, the engine:

  1. Deconstructs the Query: Tokenizes and identifies core chemical compounds, wellness goals, and formats.
  2. Performs Core Mapping: Translates long-tail queries into arrays of core database tags (e.g. mapping NMN and longevity to our white-backed resveratrol clinical sets; mapping proprietary blends to standard magnesium and vitamin D datasets).
  3. Retrieves and Aggregates: Queries the knowledge base using Prisma's powerful hasSome array comparator to pull all relevant medical abstracts in one transaction.

This optimization successfully unlocked 100% of our high-value content pipeline, allowing us to roll out 30 new deep evidence pages!

The New 30 Evidence-Based Resource Directory

Here is the complete, interlinked directory of our newly published, clinically verified supplement resources. Every page features direct studies, dosage guides, and structural markup for AI crawlers:

1. The Magnesium & Sleep Matrix

Magnesium remains the most misunderstood mineral on the market. We've compiled 8 guides to help you find the correct chemical forms, calculate actual elemental yield, and configure your evening stack:

2. Probiotics & Gut Barrier Integrity

CFU inflation is one of the most prominent label frauds. Our 6 new resources investigate the gut microbiome with scientific rigor:

3. Collagen, Skin Aging & Structural Integrity

Not all collagen is created equal. We break down the differences between hydrolyzation types, peptides, and oral absorption:

4. Immune Defense & Micronutrient Synergy

Marketing claims during cold seasons are notoriously overhyped. These 5 guides demonstrate how to create science-backed immune protocols:

5. Longevity, Cellular Energy & Nootropics

Our final group dives into the cutting edge of cellular biology, mitochondria support, and cognitive enhancers:

Moving Forward: Our Open Commitment

By mapping clinical databases directly to modern long-tail content hubs, we are building a fairer, cleaner wellness internet. We commit to keeping all our source data open, citing original clinical trials, and building our technology in the public eye.

If you have a supplement label you want parsed or a proprietary blend you want decoded, submit it directly to our database! Let's build a more transparent future together.

← All GuidesSupplement Reference →