{"id":31547,"date":"2026-01-30T14:55:38","date_gmt":"2026-01-30T22:55:38","guid":{"rendered":"https:\/\/www.pingcap.com\/?p=31547"},"modified":"2026-02-02T14:00:20","modified_gmt":"2026-02-02T22:00:20","slug":"privacy-first-ai-building-voice-to-text-app-tidb-claude","status":"publish","type":"post","link":"https:\/\/www.pingcap.com\/ko\/blog\/privacy-first-ai-building-voice-to-text-app-tidb-claude\/","title":{"rendered":"How to Build a Voice-to-Text App That Learns Your Style (Without Storing Your Words)"},"content":{"rendered":"\n<p>I&#8217;m a fast talker, but standard tools treat every platform like a dry JIRA ticket. To fix this, I dived into Chrome extension development to create <a href=\"https:\/\/app.parallellives.ai\/speak-it\">Speak It<\/a>: a voice-to-text app that learns your style without recording your secrets.<\/p>\n\n\n\n<p>Using privacy-first AI, the system maps a &#8220;fingerprint&#8221; of your speech\u2014focusing on formality and sentence length\u2014rather than storing raw content. Powered by <a href=\"https:\/\/docs.pingcap.com\/tidbcloud\/vector-search-overview\/\">TiDB vector search<\/a>, it delivers personalized formatting that satisfies even the pickiest enterprise legal teams by ensuring data is never harvested.<\/p>\n\n\n\n<p>In this blog, I&#8217;ll break down how to build a transcription tool that adapts your voice to any platform\u2014from Slack to Gmail\u2014while keeping your data completely off the server. You\u2019ll see the full technical stack as well as the &#8220;statistical fingerprinting&#8221; logic used to learn personal writing styles without ever storing an actual message.<\/p>\n\n\n<div class=\"ub-advanced-video-container wp-block-ub-advanced-video\" id=\"ub-advanced-video-8ef48007-ff59-4274-ad69-9bedb664f1f0\" ><div class=\"ub-advanced-video-embed ub-advanced-video-autofit-youtube\"><iframe loading=\"lazy\" width=\"1280\" height=\"720\" src=\"\/\/www.youtube.com\/embed\/qPYT4bi0vac\" frameborder=\"0\" allow=\"accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share\" referrerpolicy=\"strict-origin-when-cross-origin\" allowfullscreen><\/iframe><\/div><\/div>\n\n\n<h2 class=\"wp-block-heading\" id=\"0-the-technical-stack-tidb-claude-and-deepgram\"><span class=\"ez-toc-section\" id=\"The_Technical_Stack_TiDB_Claude_and_Deepgram\"><\/span>The Technical Stack: TiDB, Claude, and Deepgram<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p>Here&#8217;s what I used and why:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Chrome Extension<\/strong>: The app needs to work on any website, not just one platform. A browser extension was the only way to inject a mic button into Gmail, Slack, Notion, Twitter, and everywhere else.<\/li>\n\n\n\n<li><strong><a href=\"https:\/\/developer.mozilla.org\/en-US\/docs\/Web\/API\/Web_Speech_API\">Web Speech API<\/a> + <a href=\"https:\/\/developers.deepgram.com\/home\">Deepgram<\/a><\/strong>: Chrome and Edge support the Web Speech API for free. For browsers that don&#8217;t (Arc, Safari, Firefox), I fall back to Deepgram&#8217;s streaming API. This keeps costs low for most users while maintaining broad compatibility.<\/li>\n\n\n\n<li><strong><a href=\"https:\/\/docs.pingcap.com\/tidbcloud\/select-cluster-tier\/#starter\">TiDB Cloud Starter<\/a><\/strong>: I didn&#8217;t want to run two databases (one for normal data and one for vectors). Fortunately TiDB can handle both vectors and business data all in one database. It&#8217;s also MySQL-compatible, which means I could stick to what I already know AND it scales to zero when idle so I&#8217;m not paying for unused capacity.<\/li>\n\n\n\n<li><strong>Claude Sonnet 4<\/strong>: I use Claude Sonnet 4 as the formatting engine. It takes raw transcripts and reformats them based on context and style instructions. It&#8217;s great because Sonnet follows constraints well without over-editing (which is extremely important in this context).<\/li>\n\n\n\n<li><strong>OpenAI Embeddings<\/strong>: For embeddings, I use text-embedding-3-small with OpenAI. It generates vector representations of writing style samples. These power the similarity matching for style clustering.<\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"1-the-architecture-personalization-without-storing-user-content\"><span class=\"ez-toc-section\" id=\"The_Architecture_Personalization_Without_Storing_User_Content\"><\/span><strong>The Architecture<\/strong>: Personalization Without Storing User Content<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p>Here&#8217;s how data flows through the system:<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>&#91;User speaks] \n      \u2193\n&#91;Deepgram \/ Web Speech API]\n      \u2193\n&#91;Raw transcript]\n      \u2193\n&#91;Context detection: Gmail? Slack? Twitter?]\n      \u2193\n&#91;Fetch style profile from TiDB]\n      \u2193\n&#91;Claude formats transcript using style + context]\n      \u2193\n&#91;User accepts or rejects suggestion]\n      \u2193\n&#91;Extract stats from accepted text]\n      \u2193\n&#91;Update style profile in TiDB]\n      \u2193\n&#91;Generate embedding for similarity matching]<\/code><\/pre>\n\n\n\n<p>The key architectural decision was storing stats, not content. Here&#8217;s what goes into a style profile:<\/p>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><tbody><tr><td><strong>Field<\/strong><\/td><td><strong>Type<\/strong><\/td><td><strong>Example<\/strong><\/td><\/tr><tr><td>avg_sentence_length<\/td><td>float<\/td><td>14.2<\/td><\/tr><tr><td>formality_score<\/td><td>float (0-1)<\/td><td>0.35<\/td><\/tr><tr><td>uses_contractions<\/td><td>boolean<\/td><td>true<\/td><\/tr><tr><td>greetings<\/td><td>JSON array<\/td><td>[&#8220;Hey&#8221;, &#8220;Hi there&#8221;]<\/td><\/tr><tr><td>signoffs<\/td><td>JSON array<\/td><td>[&#8220;Thanks&#8221;, &#8220;Cheers&#8221;]<\/td><\/tr><tr><td>top_phrases<\/td><td>JSON array<\/td><td>[&#8220;sounds good&#8221;, &#8220;let me know&#8221;]<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<p>None of this is the actual message. It&#8217;s a fingerprint of how you write, not what you write.<\/p>\n\n\n\n<p>Enterprise customers won&#8217;t touch a tool that stores their internal communications. This constraint shaped every design decision.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"2-implementing-real-time-context-detection-for-gmail-slack-and-x\"><span class=\"ez-toc-section\" id=\"Implementing_Real-Time_Context_Detection_for_Gmail_Slack_and_X\"><\/span>Implementing Real-Time Context Detection for Gmail, Slack, and X<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p>Different platforms have different norms. For example, LinkedIn tends to be much more formal compared to X. And a Slack message shouldn&#8217;t read like an email. So the first thing I did was figure out where the user would be typing.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"3-context-detection\">Context Detection<\/h3>\n\n\n\n<p>The extension matches the current URL against known patterns, then looks for platform-specific DOM selectors to find the active text field:<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>const CONTEXT_PATTERNS = {\n  email: {\n    urls: &#91;\/mail\\.google\\.com\/, \/outlook\\.live\\.com\/, \/outlook\\.office\\.com\/],\n    selectors: &#91;\n      '&#91;aria-label=\"Message Body\"]',\n      '&#91;role=\"textbox\"]&#91;aria-multiline=\"true\"]',\n      'div&#91;contenteditable=\"true\"]&#91;g_editable=\"true\"]',\n    ],\n  },\n  slack: {\n    urls: &#91;\/\\.slack\\.com\/],\n    selectors: &#91;\n      '&#91;data-qa=\"message_input\"]',\n      '.ql-editor',\n      '&#91;contenteditable=\"true\"]&#91;data-message-input]',\n    ],\n  },\n  twitter: {\n    urls: &#91;\/twitter\\.com\/, \/x\\.com\/],\n    selectors: &#91;\n      '&#91;data-testid=\"tweetTextarea_0\"]',\n      '&#91;role=\"textbox\"]&#91;data-testid]',\n    ],\n  },\n  \/\/ ... 20+ contexts total\n};<\/code><\/pre>\n\n\n\n<p>This detection runs before any formatting happens. The detected context determines both how Claude formats the text and what platform-specific instructions it receives.<\/p>\n\n\n\n<p>For example, X (formerly Twitter) formatting keeps things brief and removes formal greetings while email formatting preserves sign-offs and adds paragraph breaks. And Slack sits somewhere in between.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"4-designing-a-privacy-focused-style-profile-schema-in-tidb\">Designing a Privacy-Focused Style Profile Schema in TiDB<\/h3>\n\n\n\n<p>The style profile lives in <a href=\"https:\/\/www.pingcap.com\/tidb\/\">TiDB<\/a>. Here&#8217;s the table structure:<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>CREATE TABLE user_style_profiles (\n  user_id VARCHAR(255) PRIMARY KEY,\n  avg_sentence_length FLOAT DEFAULT 12,\n  formality_score FLOAT DEFAULT 0.5,\n  uses_contractions BOOLEAN DEFAULT TRUE,\n  top_phrases JSON,\n  greetings JSON,\n  signoffs JSON,\n  created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP,\n  updated_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP\n);<\/code><\/pre>\n\n\n\n<p>Notice there&#8217;s no <code>message_content<\/code> column. We&#8217;re storing how you write, not what you write.<\/p>\n\n\n\n<p>The <code>formality_score<\/code> ranges from 0 (very casual) to 1 (very formal). This gets calculated from signals like sentence length, punctuation patterns, and word choice. Someone who writes &#8220;Hey! Quick question, can u send that over?&#8221; scores lower than someone who writes &#8220;Good afternoon. I wanted to follow up regarding the materials.&#8221;<\/p>\n\n\n\n<p>Fetching a profile is a simple query:<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>async function getUserStyleProfile(userId: string): Promise&lt;StyleProfile | null&gt; {\n  const &#91;rows] = await connection.execute&lt;RowDataPacket&#91;]&gt;(\n    `SELECT avg_sentence_length, formality_score, uses_contractions,\n            top_phrases, greetings, signoffs\n     FROM user_style_profiles WHERE user_id = ?`,\n    &#91;userId]\n  );\n  \n  if (rows.length === 0) return null;\n  \n  const row = rows&#91;0];\n  return {\n    avg_sentence_length: row.avg_sentence_length || 12,\n    formality_score: row.formality_score || 0.5,\n    uses_contractions: row.uses_contractions !== false,\n    top_phrases: row.top_phrases ? JSON.parse(row.top_phrases) : &#91;],\n    greetings: row.greetings ? JSON.parse(row.greetings) : &#91;\"Hey\"],\n    signoffs: row.signoffs ? JSON.parse(row.signoffs) : &#91;\"Thanks\"],\n  };\n}<\/code><\/pre>\n\n\n\n<p>New users get sensible defaults. The profile evolves as they accept or reject formatting suggestions.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"5-prompt-engineering-converting-style-statistics-into-claude-instructions\">Prompt Engineering: Converting Style Statistics into Claude Instructions<\/h3>\n\n\n\n<p>The style profile turns into prompt instructions. Claude doesn&#8217;t see historical messages, it sees constraints.<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>function buildStylePrompt(profile: StyleProfile | null, context: string): string {\n  if (!profile) {\n    return `Format this transcript for ${context}. Keep it natural and conversational.`;\n  }\n\n  const formality = profile.formality_score &gt; 0.7 ? \"formal\" :\n                    profile.formality_score &lt; 0.3 ? \"casual\" : \"balanced\";\n\n  const contractionNote = profile.uses_contractions\n    ? \"Use contractions naturally (don't, won't, can't).\"\n    : \"Minimize contractions for a more formal tone.\";\n\n  const greetingNote = profile.greetings.length &gt; 0\n    ? `Preferred greetings: ${profile.greetings.slice(0, 3).join(\", \")}`\n    : \"\";\n\n  const signoffNote = profile.signoffs.length &gt; 0\n    ? `Preferred sign-offs: ${profile.signoffs.slice(0, 3).join(\", \")}`\n    : \"\";\n\n  return `Format this transcript for ${context}.\n\nUser's writing style:\n- Tone: ${formality}\n- Average sentence length: ~${Math.round(profile.avg_sentence_length)} words\n- ${contractionNote}\n${greetingNote ? `- ${greetingNote}` : \"\"}\n${signoffNote ? `- ${signoffNote}` : \"\"}\n\nRules:\n1. ONLY add punctuation and paragraph breaks\n2. Remove filler words: um, uh, like, basically, you know\n3. Keep EVERY other word exactly as they said it\n4. Do NOT rewrite, rephrase, or \"clean up\" their language`;\n}<\/code><\/pre>\n\n\n\n<p>The rules at the bottom are critical. Without them, Claude will &#8220;improve&#8221; the user&#8217;s words. But people don&#8217;t want their voice replaced, they just want it cleaned up. There&#8217;s a difference.<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"698\" src=\"https:\/\/static.pingcap.com\/files\/2026\/01\/30145140\/image-5-1024x698.png\" alt=\"Privacy-first AI that keeps a user's words without replace the voice.\" class=\"wp-image-31568\" srcset=\"https:\/\/static.pingcap.com\/files\/2026\/01\/30145140\/image-5-1024x698.png 1024w, https:\/\/static.pingcap.com\/files\/2026\/01\/30145140\/image-5-300x205.png 300w, https:\/\/static.pingcap.com\/files\/2026\/01\/30145140\/image-5-768x524.png 768w, https:\/\/static.pingcap.com\/files\/2026\/01\/30145140\/image-5.png 1280w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<p>Each context also gets platform-specific instructions:<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>function getContextInstructions(context: string): string {\n  switch (context) {\n    case \"email\":\n      return `Email format:\n- Add punctuation and paragraph breaks\n- Keep their exact words\n- Add sign-off if missing`;\n\n    case \"slack\":\n      return `Slack format:\n- Keep it brief and casual\n- No formal greetings needed\n- Okay to use shorter sentences`;\n\n    case \"twitter\":\n      return `Twitter\/X format:\n- Add punctuation only\n- Keep their exact words\n- If over 280 characters, don't trim`;\n    \n    \/\/ ... more contexts\n  }\n}<\/code><\/pre>\n\n\n\n<p>The combination of style profile and context instructions gives Claude enough guidance to format appropriately without overstepping.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"6-the-learning-loop-using-weighted-averages-for-style-adaptation\">The Learning Loop: Using Weighted Averages for Style Adaptation<\/h3>\n\n\n\n<p>Here&#8217;s the part I&#8217;m still iterating on.<\/p>\n\n\n\n<p>When a user accepts or rejects a format suggestion, I want to update their profile. The naive approach was to just overwrite the stats with the new sample.<\/p>\n\n\n\n<p>But, that was wrong.<\/p>\n\n\n\n<p>If someone has been using the app for months and their profile reflects hundreds of accepted formats, a single new sample shouldn&#8217;t dramatically shift their stats. New samples need to have less influence as the profile matures.<\/p>\n\n\n\n<p>The solution is weighted averaging. Each new sample contributes a fraction to the running average, with that fraction decreasing over time:<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>function updateStyleProfile(\n  existingProfile: StyleProfile,\n  newStats: TextStats,\n  sampleCount: number\n): StyleProfile {\n  \/\/ Weight decreases as sample count increases\n  \/\/ First sample: 100% weight. 100th sample: ~1% weight.\n  const weight = 1 \/ (sampleCount + 1);\n  \n  return {\n    avg_sentence_length: \n      existingProfile.avg_sentence_length * (1 - weight) + \n      newStats.avg_sentence_length * weight,\n    formality_score:\n      existingProfile.formality_score * (1 - weight) +\n      calculateFormality(newStats) * weight,\n    \/\/ ... other fields\n  };\n}<\/code><\/pre>\n\n\n\n<p>For phrases, greetings, and signoffs, I track frequency counts rather than just presence. A greeting you use once shouldn&#8217;t rank the same as one you use constantly.<\/p>\n\n\n\n<p>I&#8217;m also generating embeddings for each accepted format:<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>const embeddingResponse = await openai.embeddings.create({\n  model: \"text-embedding-3-small\",\n  input: `Sentence length: ${stats.avg_sentence_length}. ` +\n         `Formality: ${stats.formality_score}. ` +\n         `Context: ${context}. ` +\n         `Contractions: ${stats.uses_contractions}`,\n});\nconst styleEmbedding = embeddingResponse.data&#91;0].embedding;<\/code><\/pre>\n\n\n\n<p>The idea here is to cluster similar writing styles together. Users who write like you might have formatting preferences you&#8217;d also like. But I&#8217;ll be honest: this piece isn&#8217;t fully wired up yet. I&#8217;m generating the embeddings but not querying them for recommendations.<\/p>\n\n\n\n<p>That&#8217;s the next iteration.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"7-the-result-a-cross-platform-privacy-first-voice-to-text-app\"><span class=\"ez-toc-section\" id=\"The_Result_A_Cross-Platform_Privacy-First_Voice-to-Text_App\"><\/span>The Result: A Cross-Platform, Privacy-First Voice-to-Text App<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p>What works today:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Voice-to-text on 20+ platforms (Gmail, Slack, Notion, Twitter, LinkedIn, GitHub, and more)<\/li>\n\n\n\n<li>Automatic context detection: no manual switching<\/li>\n\n\n\n<li>Style profiles that influence formatting output<\/li>\n\n\n\n<li>Privacy-first design which means statistics only, no content stored<\/li>\n<\/ul>\n\n\n\n<p>What&#8217;s next:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Vector similarity for style clustering (&#8220;users who write like you prefer&#8230;&#8221;)<\/li>\n\n\n\n<li>Refined feedback loop for profile updates<\/li>\n\n\n\n<li>Multi-language support beyond English<\/li>\n\n\n\n<li>Browser support expansion (Firefox add-on)<\/li>\n<\/ul>\n\n\n\n<p>You can find the code to the free version <a href=\"https:\/\/github.com\/RealChrisSean\/Speak-It\">here<\/a>.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"8-open-source-and-getting-started-build-your-own-transcription-tool\"><span class=\"ez-toc-section\" id=\"Open_Source_and_Getting_Started_Build_Your_Own_Transcription_Tool\"><\/span>Open Source and Getting Started: Build Your Own Transcription Tool<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p>The main insight from building this app: personalization doesn&#8217;t require surveillance. You can learn patterns without learning secrets. Statistical fingerprints give you enough signal to customize behavior while keeping actual content out of your database entirely.<\/p>\n\n\n\n<p>For enterprise use cases where privacy is non-negotiable, this approach opens doors that content-based learning keeps closed.<\/p>\n\n\n\n<p>If you want to build something similar, <a href=\"https:\/\/tidbcloud.com\/free-trial\/?__hstc=86493575.783064bfcc857ae1a573df16c96a21a4.1767977986672.1769797317712.1769809216084.83&amp;__hssc=86493575.4.1769809216084&amp;__hsfp=ef5d7ef781d92d519fb04a5267e98d6c&amp;_gl=1*1bp8hd*_gcl_au*NzgzNDI4MDk1LjE3Njc5ODI1NzU.*_ga*MjUyOTQyMTU0LjE3Njc5Nzc5ODQ.*_ga_3JVXJ41175*czE3Njk4MDc4NTYkbzkxJGcxJHQxNzY5ODEwMzg5JGo1OCRsMCRoMjA3NzM3NDkzMA..*_ga_ZEL0RNV6R2*czE3Njk4MDc4NTYkbzgwJGcxJHQxNzY5ODEwMzg5JGo2MCRsMCRoMA..*_ga_9FRXHHPYVY*czE3Njk4MDgwODQkbzkwJGcxJHQxNzY5ODEwMzg5JGo2MCRsMCRoMA..&amp;website_referrer_url=https:\/\/pingcap.zoom.us\/\">TiDB Cloud Starter<\/a> gives you enough runway to experiment. The combination of relational tables (for user profiles) and vector search (for style similarity) in one database simplified my architecture significantly.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>I&#8217;m a fast talker, but standard tools treat every platform like a dry JIRA ticket. To fix this, I dived into Chrome extension development to create Speak It: a voice-to-text app that learns your style without recording your secrets. Using privacy-first AI, the system maps a &#8220;fingerprint&#8221; of your speech\u2014focusing on formality and sentence length\u2014rather [&hellip;]<\/p>\n","protected":false},"author":324,"featured_media":31561,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"ub_ctt_via":"","footnotes":""},"categories":[436],"tags":[138,253,111,297,450],"class_list":["post-31547","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-tutorial","tag-ai","tag-security","tag-tidb","tag-vector-search","tag-voice-to-text"],"acf":[],"featured_image_src":"https:\/\/static.pingcap.com\/files\/2026\/01\/30144816\/tidb_feature_1800x600-1-5.png","author_info":{"display_name":"Chris Dabatos","author_link":"https:\/\/www.pingcap.com\/ko\/blog\/author\/chris-dabatos\/"},"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v26.9 - https:\/\/yoast.com\/product\/yoast-seo-wordpress\/ -->\n<title>Privacy-First AI: Building a Voice-to-Text App That Learns Style<\/title>\n<meta name=\"description\" content=\"Build a privacy-first, voice-to-text AI app using TiDB and Claude that learns writing styles without storing user messages.\" \/>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.pingcap.com\/ko\/blog\/privacy-first-ai-building-voice-to-text-app-tidb-claude\/\" \/>\n<meta property=\"og:locale\" content=\"ko_KR\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Privacy-First AI: Building a Voice-to-Text App That Learns Style\" \/>\n<meta property=\"og:description\" content=\"Build a privacy-first, voice-to-text AI app using TiDB and Claude that learns writing styles without storing user messages.\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.pingcap.com\/ko\/blog\/privacy-first-ai-building-voice-to-text-app-tidb-claude\/\" \/>\n<meta property=\"og:site_name\" content=\"TiDB\" \/>\n<meta property=\"article:publisher\" content=\"https:\/\/facebook.com\/pingcap2015\" \/>\n<meta property=\"article:published_time\" content=\"2026-01-30T22:55:38+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2026-02-02T22:00:20+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/static.pingcap.com\/files\/2026\/01\/30144834\/tidb_1200x627-2-2.png\" \/>\n\t<meta property=\"og:image:width\" content=\"2400\" \/>\n\t<meta property=\"og:image:height\" content=\"1254\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/png\" \/>\n<meta name=\"author\" content=\"Chris Dabatos\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:image\" content=\"https:\/\/static.pingcap.com\/files\/2026\/01\/30144850\/tidb_twitter_1600x900-3-4.png\" \/>\n<meta name=\"twitter:creator\" content=\"@PingCAP\" \/>\n<meta name=\"twitter:site\" content=\"@PingCAP\" \/>\n<meta name=\"twitter:label1\" content=\"Written by\" \/>\n\t<meta name=\"twitter:data1\" content=\"Chris Dabatos\" \/>\n\t<meta name=\"twitter:label2\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"8\ubd84\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\/\/www.pingcap.com\/blog\/privacy-first-ai-building-voice-to-text-app-tidb-claude\/#article\",\"isPartOf\":{\"@id\":\"https:\/\/www.pingcap.com\/blog\/privacy-first-ai-building-voice-to-text-app-tidb-claude\/\"},\"author\":{\"name\":\"Chris Dabatos\",\"@id\":\"https:\/\/www.pingcap.com\/#\/schema\/person\/4d7ecdb90868256414855723f838c9e0\"},\"headline\":\"How to Build a Voice-to-Text App That Learns Your Style (Without Storing Your Words)\",\"datePublished\":\"2026-01-30T22:55:38+00:00\",\"dateModified\":\"2026-02-02T22:00:20+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\/\/www.pingcap.com\/blog\/privacy-first-ai-building-voice-to-text-app-tidb-claude\/\"},\"wordCount\":1167,\"publisher\":{\"@id\":\"https:\/\/www.pingcap.com\/#organization\"},\"image\":{\"@id\":\"https:\/\/www.pingcap.com\/blog\/privacy-first-ai-building-voice-to-text-app-tidb-claude\/#primaryimage\"},\"thumbnailUrl\":\"https:\/\/static.pingcap.com\/files\/2026\/01\/30144816\/tidb_feature_1800x600-1-5.png\",\"keywords\":[\"AI\",\"Security\",\"TiDB\",\"Vector Search\",\"Voice-to-Text\"],\"articleSection\":[\"Tutorial\"],\"inLanguage\":\"ko-KR\"},{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.pingcap.com\/blog\/privacy-first-ai-building-voice-to-text-app-tidb-claude\/\",\"url\":\"https:\/\/www.pingcap.com\/blog\/privacy-first-ai-building-voice-to-text-app-tidb-claude\/\",\"name\":\"Privacy-First AI: Building a Voice-to-Text App That Learns Style\",\"isPartOf\":{\"@id\":\"https:\/\/www.pingcap.com\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\/\/www.pingcap.com\/blog\/privacy-first-ai-building-voice-to-text-app-tidb-claude\/#primaryimage\"},\"image\":{\"@id\":\"https:\/\/www.pingcap.com\/blog\/privacy-first-ai-building-voice-to-text-app-tidb-claude\/#primaryimage\"},\"thumbnailUrl\":\"https:\/\/static.pingcap.com\/files\/2026\/01\/30144816\/tidb_feature_1800x600-1-5.png\",\"datePublished\":\"2026-01-30T22:55:38+00:00\",\"dateModified\":\"2026-02-02T22:00:20+00:00\",\"description\":\"Build a privacy-first, voice-to-text AI app using TiDB and Claude that learns writing styles without storing user messages.\",\"breadcrumb\":{\"@id\":\"https:\/\/www.pingcap.com\/blog\/privacy-first-ai-building-voice-to-text-app-tidb-claude\/#breadcrumb\"},\"inLanguage\":\"ko-KR\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.pingcap.com\/blog\/privacy-first-ai-building-voice-to-text-app-tidb-claude\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"ko-KR\",\"@id\":\"https:\/\/www.pingcap.com\/blog\/privacy-first-ai-building-voice-to-text-app-tidb-claude\/#primaryimage\",\"url\":\"https:\/\/static.pingcap.com\/files\/2026\/01\/30144816\/tidb_feature_1800x600-1-5.png\",\"contentUrl\":\"https:\/\/static.pingcap.com\/files\/2026\/01\/30144816\/tidb_feature_1800x600-1-5.png\",\"width\":3600,\"height\":1200},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.pingcap.com\/blog\/privacy-first-ai-building-voice-to-text-app-tidb-claude\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\/\/www.pingcap.com\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"How to Build a Voice-to-Text App That Learns Your Style (Without Storing Your Words)\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.pingcap.com\/#website\",\"url\":\"https:\/\/www.pingcap.com\/\",\"name\":\"TiDB\",\"description\":\"TiDB | SQL at Scale\",\"publisher\":{\"@id\":\"https:\/\/www.pingcap.com\/#organization\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.pingcap.com\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"ko-KR\"},{\"@type\":\"Organization\",\"@id\":\"https:\/\/www.pingcap.com\/#organization\",\"name\":\"PingCAP\",\"url\":\"https:\/\/www.pingcap.com\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"ko-KR\",\"@id\":\"https:\/\/www.pingcap.com\/#\/schema\/logo\/image\/\",\"url\":\"https:\/\/static.pingcap.com\/files\/2021\/11\/pingcap-logo.png\",\"contentUrl\":\"https:\/\/static.pingcap.com\/files\/2021\/11\/pingcap-logo.png\",\"width\":811,\"height\":232,\"caption\":\"PingCAP\"},\"image\":{\"@id\":\"https:\/\/www.pingcap.com\/#\/schema\/logo\/image\/\"},\"sameAs\":[\"https:\/\/facebook.com\/pingcap2015\",\"https:\/\/x.com\/PingCAP\",\"https:\/\/linkedin.com\/company\/pingcap\",\"https:\/\/youtube.com\/channel\/UCuq4puT32DzHKT5rU1IZpIA\"]},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.pingcap.com\/#\/schema\/person\/4d7ecdb90868256414855723f838c9e0\",\"name\":\"Chris Dabatos\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"ko-KR\",\"@id\":\"https:\/\/www.pingcap.com\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/static.pingcap.com\/files\/2022\/10\/17234942\/avatar.jpg\",\"contentUrl\":\"https:\/\/static.pingcap.com\/files\/2022\/10\/17234942\/avatar.jpg\",\"caption\":\"Chris Dabatos\"},\"description\":\"Developer Advocate\",\"url\":\"https:\/\/www.pingcap.com\/ko\/blog\/author\/chris-dabatos\/\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Privacy-First AI: Building a Voice-to-Text App That Learns Style","description":"Build a privacy-first, voice-to-text AI app using TiDB and Claude that learns writing styles without storing user messages.","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.pingcap.com\/ko\/blog\/privacy-first-ai-building-voice-to-text-app-tidb-claude\/","og_locale":"ko_KR","og_type":"article","og_title":"Privacy-First AI: Building a Voice-to-Text App That Learns Style","og_description":"Build a privacy-first, voice-to-text AI app using TiDB and Claude that learns writing styles without storing user messages.","og_url":"https:\/\/www.pingcap.com\/ko\/blog\/privacy-first-ai-building-voice-to-text-app-tidb-claude\/","og_site_name":"TiDB","article_publisher":"https:\/\/facebook.com\/pingcap2015","article_published_time":"2026-01-30T22:55:38+00:00","article_modified_time":"2026-02-02T22:00:20+00:00","og_image":[{"width":2400,"height":1254,"url":"https:\/\/static.pingcap.com\/files\/2026\/01\/30144834\/tidb_1200x627-2-2.png","type":"image\/png"}],"author":"Chris Dabatos","twitter_card":"summary_large_image","twitter_image":"https:\/\/static.pingcap.com\/files\/2026\/01\/30144850\/tidb_twitter_1600x900-3-4.png","twitter_creator":"@PingCAP","twitter_site":"@PingCAP","twitter_misc":{"Written by":"Chris Dabatos","Est. reading time":"8\ubd84"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/www.pingcap.com\/blog\/privacy-first-ai-building-voice-to-text-app-tidb-claude\/#article","isPartOf":{"@id":"https:\/\/www.pingcap.com\/blog\/privacy-first-ai-building-voice-to-text-app-tidb-claude\/"},"author":{"name":"Chris Dabatos","@id":"https:\/\/www.pingcap.com\/#\/schema\/person\/4d7ecdb90868256414855723f838c9e0"},"headline":"How to Build a Voice-to-Text App That Learns Your Style (Without Storing Your Words)","datePublished":"2026-01-30T22:55:38+00:00","dateModified":"2026-02-02T22:00:20+00:00","mainEntityOfPage":{"@id":"https:\/\/www.pingcap.com\/blog\/privacy-first-ai-building-voice-to-text-app-tidb-claude\/"},"wordCount":1167,"publisher":{"@id":"https:\/\/www.pingcap.com\/#organization"},"image":{"@id":"https:\/\/www.pingcap.com\/blog\/privacy-first-ai-building-voice-to-text-app-tidb-claude\/#primaryimage"},"thumbnailUrl":"https:\/\/static.pingcap.com\/files\/2026\/01\/30144816\/tidb_feature_1800x600-1-5.png","keywords":["AI","Security","TiDB","Vector Search","Voice-to-Text"],"articleSection":["Tutorial"],"inLanguage":"ko-KR"},{"@type":"WebPage","@id":"https:\/\/www.pingcap.com\/blog\/privacy-first-ai-building-voice-to-text-app-tidb-claude\/","url":"https:\/\/www.pingcap.com\/blog\/privacy-first-ai-building-voice-to-text-app-tidb-claude\/","name":"Privacy-First AI: Building a Voice-to-Text App That Learns Style","isPartOf":{"@id":"https:\/\/www.pingcap.com\/#website"},"primaryImageOfPage":{"@id":"https:\/\/www.pingcap.com\/blog\/privacy-first-ai-building-voice-to-text-app-tidb-claude\/#primaryimage"},"image":{"@id":"https:\/\/www.pingcap.com\/blog\/privacy-first-ai-building-voice-to-text-app-tidb-claude\/#primaryimage"},"thumbnailUrl":"https:\/\/static.pingcap.com\/files\/2026\/01\/30144816\/tidb_feature_1800x600-1-5.png","datePublished":"2026-01-30T22:55:38+00:00","dateModified":"2026-02-02T22:00:20+00:00","description":"Build a privacy-first, voice-to-text AI app using TiDB and Claude that learns writing styles without storing user messages.","breadcrumb":{"@id":"https:\/\/www.pingcap.com\/blog\/privacy-first-ai-building-voice-to-text-app-tidb-claude\/#breadcrumb"},"inLanguage":"ko-KR","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.pingcap.com\/blog\/privacy-first-ai-building-voice-to-text-app-tidb-claude\/"]}]},{"@type":"ImageObject","inLanguage":"ko-KR","@id":"https:\/\/www.pingcap.com\/blog\/privacy-first-ai-building-voice-to-text-app-tidb-claude\/#primaryimage","url":"https:\/\/static.pingcap.com\/files\/2026\/01\/30144816\/tidb_feature_1800x600-1-5.png","contentUrl":"https:\/\/static.pingcap.com\/files\/2026\/01\/30144816\/tidb_feature_1800x600-1-5.png","width":3600,"height":1200},{"@type":"BreadcrumbList","@id":"https:\/\/www.pingcap.com\/blog\/privacy-first-ai-building-voice-to-text-app-tidb-claude\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/www.pingcap.com\/"},{"@type":"ListItem","position":2,"name":"How to Build a Voice-to-Text App That Learns Your Style (Without Storing Your Words)"}]},{"@type":"WebSite","@id":"https:\/\/www.pingcap.com\/#website","url":"https:\/\/www.pingcap.com\/","name":"\ud2f0DB","description":"TiDB | SQL at Scale","publisher":{"@id":"https:\/\/www.pingcap.com\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.pingcap.com\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"ko-KR"},{"@type":"Organization","@id":"https:\/\/www.pingcap.com\/#organization","name":"PingCAP","url":"https:\/\/www.pingcap.com\/","logo":{"@type":"ImageObject","inLanguage":"ko-KR","@id":"https:\/\/www.pingcap.com\/#\/schema\/logo\/image\/","url":"https:\/\/static.pingcap.com\/files\/2021\/11\/pingcap-logo.png","contentUrl":"https:\/\/static.pingcap.com\/files\/2021\/11\/pingcap-logo.png","width":811,"height":232,"caption":"PingCAP"},"image":{"@id":"https:\/\/www.pingcap.com\/#\/schema\/logo\/image\/"},"sameAs":["https:\/\/facebook.com\/pingcap2015","https:\/\/x.com\/PingCAP","https:\/\/linkedin.com\/company\/pingcap","https:\/\/youtube.com\/channel\/UCuq4puT32DzHKT5rU1IZpIA"]},{"@type":"Person","@id":"https:\/\/www.pingcap.com\/#\/schema\/person\/4d7ecdb90868256414855723f838c9e0","name":"Chris Dabatos","image":{"@type":"ImageObject","inLanguage":"ko-KR","@id":"https:\/\/www.pingcap.com\/#\/schema\/person\/image\/","url":"https:\/\/static.pingcap.com\/files\/2022\/10\/17234942\/avatar.jpg","contentUrl":"https:\/\/static.pingcap.com\/files\/2022\/10\/17234942\/avatar.jpg","caption":"Chris Dabatos"},"description":"Developer Advocate","url":"https:\/\/www.pingcap.com\/ko\/blog\/author\/chris-dabatos\/"}]}},"grav_blocks":false,"card_markup":"<a class=\"card-resource bg-white\" href=\"https:\/\/www.pingcap.com\/ko\/blog\/privacy-first-ai-building-voice-to-text-app-tidb-claude\/\"><div class=\"card-resource__image-container\"><img class=\"card-resource__image\" alt=\"tidb_feature_1800x600 (1)\" src=\"https:\/\/static.pingcap.com\/files\/2026\/01\/30144816\/tidb_feature_1800x600-1-5.png\" loading=\"lazy\" width=3600 height=1200 \/><\/div><div class=\"card-resource__content-container\"><div class=\"card-resource__content-head\"><div class=\"card-resource__category\">Tutorial<\/div><\/div><h5 class=\"card-resource__title\">How to Build a Voice-to-Text App That Learns Your Style (Without Storing Your Words)<\/h5><\/div><\/a>","_links":{"self":[{"href":"https:\/\/www.pingcap.com\/ko\/wp-json\/wp\/v2\/posts\/31547","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.pingcap.com\/ko\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.pingcap.com\/ko\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.pingcap.com\/ko\/wp-json\/wp\/v2\/users\/324"}],"replies":[{"embeddable":true,"href":"https:\/\/www.pingcap.com\/ko\/wp-json\/wp\/v2\/comments?post=31547"}],"version-history":[{"count":15,"href":"https:\/\/www.pingcap.com\/ko\/wp-json\/wp\/v2\/posts\/31547\/revisions"}],"predecessor-version":[{"id":31909,"href":"https:\/\/www.pingcap.com\/ko\/wp-json\/wp\/v2\/posts\/31547\/revisions\/31909"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.pingcap.com\/ko\/wp-json\/wp\/v2\/media\/31561"}],"wp:attachment":[{"href":"https:\/\/www.pingcap.com\/ko\/wp-json\/wp\/v2\/media?parent=31547"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.pingcap.com\/ko\/wp-json\/wp\/v2\/categories?post=31547"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.pingcap.com\/ko\/wp-json\/wp\/v2\/tags?post=31547"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}