{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreicqhg336beujv2yei7lsyk6xj2kuh7ahtq4bafkmx2mnha6w5dhfa",
    "uri": "at://did:plc:7basvtgtkukzhs5nx64oqgvp/app.bsky.feed.post/3miukboytsza2"
  },
  "description": "How I went from asking 'Why is my glucose high?' to getting personalized daily coaching that improved my time in range from 82% to 98%. Six iterations of prompt engineering that transformed generic AI into an essential health tool.",
  "path": "/ai-coach-part-3/",
  "publishedAt": "2026-04-07T00:42:26.000Z",
  "site": "https://blog.warrenweb.net",
  "tags": [
    "The Transformation",
    "System Architecture",
    "Prompt Engineering",
    "Knowledge Graph",
    "Build Your Own",
    "Teaching Your AI Coach to Cook",
    "Day Cluster"
  ],
  "textContent": "* * *\n\n**🤖 _AI Diabetes Coach_ series:**\n_— 🚀 Part 1:_ The Transformation\n_— 🏗️ Part 2:_ System Architecture\n_➔ 💭 Part 3:_ Prompt Engineering\n_— 🕸️ Part 4:_ Knowledge Graph\n_—_ 🔧 Part 5: Build Your Own\n_—_ 🍴 __ Part 6: Teaching Your AI Coach to Cook\n\n* * *\n\n## The Generic AI Problem\n\n _Why basic ChatGPT doesn't work for health coaching_\n\nWith diabetes (or any health issue), asking an AI a generic question gets you a generic answer—interesting perhaps, but rarely helpful. As they say, \"Garbage in, garbage out\".\n\nFor example, if I ask ChatGPT:\n\n> Why is my glucose high?\"\n\nI get something like this:\n\n> High blood glucose can happen for several reasons, depending on your situation (whether you have diabetes, what you’ve eaten, stress levels, etc.). Common causes include:\n>\n> **- Diabetes** - Your body can't regulate sugar properly- **Food** - Carbs raise blood sugar- **Stress and illness** - Hormones affect glucose- **Medications** - Some drugs raise blood sugar- **Lack of sleep** - Disrupts insulin sensitivity- **Inactivity** - Exercise helps lower glucose- **Dawn phenomenon** - Morning hormone surge\n> **What you can do right now:** Check your glucose againDrink waterTake a short walk if safeReview what you ate and when\n>  _If you want, tell me your glucose reading(s), when you measured (fasting, after eating, etc.), and any symptoms... and I can help you narrow down the likely cause._\n\nSo what does this tell me? Not much.\n\nIt's like asking \"Why is the sky blue?\" without any context about where I am, what time it is, or what I'm actually seeing. The response lists every possible reason the sky _might_ appear blue, but doesn't help me understand \"my specific sky\" right now.\n\n**What I actually need** is to understand my timelines, patterns, and trends. Not generic possibilities—but specific insights about _my_ glucose, _my_ meals, _my_ insulin, and _my_ daily patterns. That's what lets me learn and adapt to improve my health.\n\nGeneric AI can't do that. But engineered prompts can.\n\n## Six Prompt Iterations\n\n _How I evolved from questions to coaching_\n\n### Iteration 1: Basic Questions\n\n _No context, generic answers_\n\n**What I tried:**\nJust asking health questions directly:\n\n  * \"What should my glucose be after breakfast?\"\n  * \"Is 156 mg/dL too high?\"\n  * \"Why did I spike to 180?\"\n\n\n\n**What I got:**\nGeneric medical ranges and boilerplate advice:\n\n  * \"Normal post-meal glucose is 140-180 mg/dL\"\n  * \"Consult your doctor about target ranges\"\n  * Spikes can be caused by many factors...\"\n\n\n\n**What was missing:**\nAny knowledge of ME:\n\n  * My Type 1 diabetes\n  * My insulin doses\n  * My meal timing\n  * My target ranges (70-160, not 140-180!)\n  * My patterns over time\n\n\n\n**What I learned:**\nAI needs MY context, not generic medical facts.\n\n### Iteration 2: Added My Diabetes Type\n\n _What changed, why it mattered_\n\n**What I tried:**\n\n  * \"I have Type 1 diabetes. Why is my glucose high?\"\n  * As someone with T1D, is 156 too high after breakfast?\n\n\n\n**What I got:**\nType 1-specific information—better than generic:\n\n  * References to insulin dependence\n  * Mentions of carb counting\n  * Tighter target ranges suggested\n\n\n\n**What was missing:**\nStill no knowledge of MY specific situation:\n\n  * My current insulin regimen\n  * My actual breakfast and timing\n  * My glucose trend (rising? stable? falling?)\n  * My exercise plans or stress levels\n\n\n\n**What I learned:**\nDiagnosis is important context, but I needed current data—not just historical facts about my condition.\n\n### Iteration 3: Added Current Data\n\n\n**What I tried:**\nIncluding real-time information:\n\n  * \"I have Type 1 diabetes. My glucose is 156 at 10 AM. I ate breakfast at 7:30 with 3u insulin. Why am I high?\"\n\n\n\n**What I got:**\nMuch more specific analysis:\n\n  * Considered insulin-to-carb ratio\n  * Looked at 2-hour post-meal timing\n  * Suggested possible carb underestimation\n  * Mentioned dawn phenomenon possibility\n\n\n\n**What was missing:**\nContext beyond this single moment:\n\n  * What did I eat for breakfast?\n  * How much did I actually carb-count?\n  * Is this high normal for me after this meal?\n  * How does today compare to yesterday?\n\n\n\n**What I learned:**\nOne data point tells a story, but patterns tell the truth.\n\n### Iteration 4: Added Daily Pattern\n\n\n**What I tried:**\nProviding my full day's data:\n\n  * All glucose readings\n  * Every meal with carb counts\n  * All insulin doses with timing\n  * Exercise or unusual events\n\n\n\n**What I got:**\nPattern recognition within the day:\n\n  * \"Your breakfast spike suggests under-dosing by ~1u\"\n  * \"Lunch was stable—that ratio worked well\"\n  * \"Evening low might be from afternoon exercise\"\n\n\n\n**What was missing:**\nHistorical comparison:\n\n  * Is today typical or unusual?\n  * Which patterns repeat across days?\n  * What changed from yesterday to today?\n\n\n\n**What I learned:**\nDaily patterns reveal cause and effect, but I needed to compare across days to see what's sustainable vs. what's a one-time event.\n\n### Iteration 5: Added Weekly Context\n\n\n**What I tried:**\nPasting several days worth of data:\n\n  * Monday through Friday glucose logs\n  * All meals and insulin doses\n  * Exercise and stress notes\n\n\n\n**What I got:**\nMulti-day pattern analysis:\n\n  * \"You consistently spike after breakfast tacos\"\n  * \"Your Friday evening lows correlate with yard work\"\n  * \"Weekend patterns differ from weekdays\"\n\n\n\n**What was missing:**\nThis was incredibly powerful... but exhausting:\n\n  * Manually copying/pasting data each time\n  * No memory between conversations\n  * Starting from scratch every session\n  * Unsustainable long-term\n\n\n\n**What I learned:**\nI needed automation and persistent memory, not just more data.\n\n### Iteration 6: The Breakthrough\n\n\n**What I tried:**\nThree game-changing elements:\n\n  1. **Claude Projects** for persistent memory across conversations\n  2. **Automated data sync** via Dexcom API (glucose) + Glooko (insulin)\n  3. **Structured daily reviews** generated from my own data\n\n\n\n**What I got:**\nTrue personalized coaching:\n\n  * Daily analysis of yesterday's patterns\n  * Comparisons to my historical trends\n  * Specific, actionable recommendations\n  * Conversational dialogue that builds on prior discussions\n  * No manual data entry required\n\n\n\n**What was missing:**\nNothing. This was the breakthrough.\n\n**What I learned:**\nThe combination of automation, memory, and structure transformed AI from \"occasionally helpful\" to \"daily essential coaching.\"\n\n**Result**\n82% → 98% time in range over 14 days. The proof that engineered prompts work.\n\n## What Makes It Work\n\n _Specific, context, examples, conversational_\n\nFour principles emerged from my iterations:\n\n### 1. Specificity over Generality\n\nDon't ask: \"What should I eat for breakfast?\"\nAsk: \"I have Type 1 diabetes, my fasting glucose is 110, I have 2 hours before a morning workout— what's a good breakfast that won't spike me during exercise?\"\n\nGeneric questions get generic answers. Specific questions get actionable guidance.\n\n### 2. Context over Instructions\n\nDon't say: \"Analyze my glucose\".\nSay: \"Here's today's glucose data, meals, and insulin doses. I spiked to 210 after lunch yesterday with the same meal but didn't today— what was different?\"\n\nContext lets AI see patterns you might miss.\n\n### 3. Examples over Explanations\n\nDon't explain: \"I want daily summaries\".\nShow: \"Here's the format I want:\n\n  * Average glucose: X\n  * Time in range: Y%\n  * Meals that worked well: Z\n  * Patterns to watch: ...\"\n\n\n\nExamples are clearer than descriptions.\n\n### 4. Conversation over Commands\n\nDon't command: \"Give me a meal plan\".\nConverse: \"Looking at my patterns this week, which meals kept me most stable?\n\nCan we build on those?\" Dialog builds understanding over time.\n\n## The Results\n\n _Real data from real days_\n\nAfter implementing Iteration 6 (automated data sync + Claude Projects + daily reviews), my diabetes control transformed:\n\n**Before engineered prompts:**\n\n  * Time in Range (Tight 70-160): 82%\n  * Inconsistent patterns\n  * Reactive management (correcting problems)\n  * Frustration with unexplained spikes and crashes\n\n\n\n**After engineered prompts:**\n\n  * Time in Range (Tight 70-160): 98%\n  * Clear pattern recognition\n  * Proactive management (preventing problems)\n  * Understanding of my specific triggers\n\n\n\n**14-day streak results:**\n\n  * Average Tight TIR: 92%\n  * Standard TIR (70-180): 97%\n  * Learned from every challenging day\n  * Recovered quickly from weekend restaurant meals\n  * Identified exercise timing effects\n  * Understood my roller coaster patterns\n\n\n\nThe difference wasn't just numbers—it was confidence. Instead of guessing why my glucose did something, I could trace cause and effect. Instead of generic advice, I got specific insights about MY body's patterns.\n\nEngineered prompts turned AI from a search engine into a coach.\n\n## Getting Started\n\n _A progressive approach for your own prompts_\n\nYou don't need to jump straight to Iteration 6. Start simple and add complexity as you learn what helps:\n\n**Week 1: Start with specificity**\n\n  * Replace \"Help me with diabetes\" with \"I have Type 1 diabetes, my fasting glucose is 110, what should I consider for breakfast?\"\n  * Add your diagnosis, current situation, and specific question\n\n\n\n**Week 2: Add context**\n\n  * Include your current data: glucose reading, timing, recent meal, insulin dose\n  * Ask about relationships: \"What might explain this spike?\"\n\n\n\n**Week 3: Show patterns**\n\n  * Share 2-3 days of data\n  * Ask for pattern recognition\n  * Notice what insights emerge from comparison\n\n\n\n**Week 4: Experiment with structure**\n\n  * Try different data formats\n  * See what Claude (or ChatGPT or Gemini) understands best\n  * Refine based on what gives useful responses\n\n\n\n**Eventually: Automate what works**\n\n  * If manually entering data gets tedious, consider automation\n  * Use Claude Projects (or ChatGPT memory) for persistence\n  * Build on what you've learned through experimentation\n\n\n\n**Remember:**\n\n  * Start where you are (no judgment!)\n  * Add one improvement at a time (iterate!)\n  * Keep what works, discard what doesn't (learn!)\n  * Share what you discover (help others!)\n\n\n\nThe goal isn't perfection—it's progress. Each iteration taught me something. Yours will too.\n\n## What's Next\n\n  * 🕸️ **Part 4:** Knowledge Graph — Giving AI persistent, queryable memory so it stops forgetting everything between sessions.\n  * 🔧**Part 5:** Build Your Own — How to start building your own AI health system, regardless of technical background. This puts it all together with AI-powered analysis and the PKM services that tie the graph to your daily workflow.\n  * 🍴 __**Part 6:** Teaching Your AI Coach to Cook — How food choices connect directly to glucose outcomes — building a recipe intelligence system that learns from your actual responses.\n\n\n\n_Start from the beginning: 🚀 Part 1:_ The Transformation\n\nThe PKM structure that organizes all this data is covered in a separate series: 📅 Day Cluster (Part 1).",
  "title": "💭 AI Diabetes Coach—AI Prompt Engineering",
  "updatedAt": "2026-04-28T12:00:16.746Z"
}