Question 1

What makes this editor's paste cleanup different from Notion or Word?

Accepted Answer

Notion and Word prioritize fidelity — they keep pasted junk. This editor rebuilds pasted content into clean structured nodes using a deterministic pipeline, so you get clean output instead of scrubbing junk by hand.

Question 2

Why is there no LLM in the paste pipeline?

Accepted Answer

Cleanup is a structure problem, not a language problem. Deterministic rules make paste instant, free, repeatable, and meaning-preserving. An LLM would be slower, cost tokens per paste, and could silently reword your content.

Question 3

How does it fix PDF copy-paste line wraps?

Accepted Answer

Line-break repair joins consecutive lines where the previous doesn't end in terminal punctuation and the next starts lowercase — with spaCy sentence-boundary fallback server-side.

Question 4

I pasted ChatGPT output with literal ** and #. Does this fix that?

Accepted Answer

Yes. markdownToHtml detects markdown literals and converts them to real formatting — headings, bold, lists, tables, code fences — with no LLM involved.

Question 5

Does the AI ever run automatically on paste?

Accepted Answer

No. AI is opt-in and user-initiated. The SmartPaste extension imports no AI client and makes no AI calls.

Question 6

Can the AI silently overwrite my text?

Accepted Answer

No. AI output streams below your selection and requires explicit confirm to apply, or cancel/undo — nothing is replaced without confirmation.

Question 7

Why TipTap/ProseMirror instead of Lexical?

Accepted Answer

ProseMirror gives schema-validated re-parsing (free sanitization), a first-class paste-handler plugin, and single-transaction undo. The FastAPI server pipeline emits Lexical JSON as a parallel output format.

Question 8

How is the app protected against prompt injection?

Accepted Answer

sanitize_for_llm runs Unicode NFKD normalization, base64 rescan, and an injection-pattern denylist. Pasted content is always wrapped as data, not instructions.

Question 9

What is Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs?

Accepted Answer

Full-stack rich text editor whose core value is deterministic paste cleanup — rules, parsers, and heuristics turn PDF wraps, Word junk HTML, ChatGPT markdown literals, and spreadsheet cells into clean TipTap nodes with zero LLM calls in the paste path. A separate opt-in AI layer adds 9 rewrite modes, SSE streaming, ghost autocomplete, and 20 AI endpoints — all previewed before apply.

Question 10

What problem does Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs solve?

Accepted Answer

Pasting text into an editor is where formatting goes to die. PDF copy-paste hard-wraps mid-sentence. Word and Google Docs inject kilobytes of mso-* classes, empty spans, and fake-bold markup. ChatGPT output pastes markdown literals (`**bold**`, `# Heading`) as plain text. Spreadsheets collapse into one run-on line.

Traditional editors (Notion, Word, Google Docs) prioritize fidelity — they keep t…

Question 11

Why was Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs built?

Accepted Answer

I kept hitting the same wall: PDF paragraphs became ragged half-sentences, Word paste carried invisible garbage, and ChatGPT output showed literal asterisks instead of formatting.

Throwing an LLM at every paste felt wrong — cleanup is structural, deterministic, and meaning-preserving. I built the cleanup as regex normalization, Office HTML sanitization, structured-document parsing, and CSV/markd…

Question 12

How does Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs work at a high level?

Accepted Answer

Built Smart Text Editor: TipTap paste pipeline with Office HTML sanitizer, structured-doc parser, CSV/markdown converters, and URL embeds — zero LLM in paste. Opt-in AI layer: 9 rewrite modes, SSE streaming, ghost autocomplete, confirm-before-apply.

Question 13

What was the business impact of Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs?

Accepted Answer

Instant deterministic paste cleanup locked by golden fixtures. 20 opt-in AI endpoints with preview-before-apply and prompt-injection guardrails. Judgment-driven design: structure problems don't need LLMs.

Question 14

What type of project is Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs?

Accepted Answer

Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs is a Full-Stack project delivered in 2026, with Solo Builder ownership across architecture and implementation.

Question 15

What was Solo Builder's role on Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs?

Accepted Answer

Built the deterministic SmartPaste pipeline — convertPastedContent, cleanHtml, parseStructuredBlocks, CSV/markdown/URL detectors, and Turndown normalization. Built the FastAPI 5-stage paste pipeline with golden-fixture pytest coverage and Lexical JSON output. Built 20 Next.js AI route handlers with SSE streaming, mode→prompt mapping, and per-feature cost tracking. Built ghost-text AiAutocomplete…

Question 16

What technologies power Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs?

Accepted Answer

Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs uses TipTap 3, ProseMirror, Next.js 15, React 19, TypeScript, Tailwind CSS, Zustand, Turndown, FastAPI, spaCy, ftfy, OpenAI, and additional production tooling.

Question 17

What AI capabilities are included in Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs?

Accepted Answer

Dhruvil Patel built Smart Text Editor — a TipTap/ProseMirror + Next.js 15 + FastAPI rich text editor whose core innovation is deterministic paste cleanup (PDF, Word, ChatGPT markdown, spreadsheets) with zero LLM calls in the paste path, plus a strictly opt-in AI layer with 9 rewrite modes, SSE streaming, ghost autocomplete, preview-before-apply, and prompt-injection guardrails.

Question 18

How does Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs relate to Dhruvil Patel's portfolio?

Accepted Answer

Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs demonstrates production AI engineering in Full-Stack, with measurable outcomes and documented architecture.

Question 19

Can I see a live demo of Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs?

Accepted Answer

Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs is documented in this case study with architecture, approach, and outcomes.

Question 20

Where is the source code for Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs?

Accepted Answer

Implementation detail is covered in the approach, architecture, and tradeoffs sections of this case study.

Question 21

What happens in approach step 1 of Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs?

Accepted Answer

Paste interception via SmartPaste TipTap extension — handlePaste reads text/plain and text/html, routes internal copies, URLs, and general content before default handler.

Question 22

How does Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs handle: Paste interception via SmartPaste TipTap extension — handlePaste reads text/pla…?

Accepted Answer

Paste interception via SmartPaste TipTap extension — handlePaste reads text/plain and text/html, routes internal copies, URLs, and general content before default handler.

Question 23

What happens in approach step 2 of Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs?

Accepted Answer

Office/Docs HTML sanitizer (cleanHtml) strips mso-*, , , fake-bold spans, empty spans, and runs.

Question 24

How does Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs handle: Office/Docs HTML sanitizer (cleanHtml) strips mso-*, , , fake-bold sp…?

Accepted Answer

Office/Docs HTML sanitizer (cleanHtml) strips mso-*, , , fake-bold spans, empty spans, and runs.

Question 25

What happens in approach step 3 of Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs?

Accepted Answer

Source disambiguation (isBrowserPlainTextHtml) detects trivial browser HTML wrapping and routes to structured parser.

Question 26

How does Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs handle: Source disambiguation (isBrowserPlainTextHtml) detects trivial browser HTML wra…?

Accepted Answer

Source disambiguation (isBrowserPlainTextHtml) detects trivial browser HTML wrapping and routes to structured parser.

Question 27

What happens in approach step 4 of Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs?

Accepted Answer

Structured-document parser (parseStructuredBlocks) promotes titles, numbered sections, bullets, and ALL-CAPS lines to proper heading/list nodes.

Question 28

How does Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs handle: Structured-document parser (parseStructuredBlocks) promotes titles, numbered se…?

Accepted Answer

Structured-document parser (parseStructuredBlocks) promotes titles, numbered sections, bullets, and ALL-CAPS lines to proper heading/list nodes.

Question 29

What happens in approach step 5 of Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs?

Accepted Answer

CSV/TSV detector (looksLikeCsv → csvToHtml) with quoted-field handling builds proper tables from spreadsheet paste.

Question 30

How does Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs handle: CSV/TSV detector (looksLikeCsv → csvToHtml) with quoted-field handling builds p…?

Accepted Answer

CSV/TSV detector (looksLikeCsv → csvToHtml) with quoted-field handling builds proper tables from spreadsheet paste.

Question 31

What happens in approach step 6 of Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs?

Accepted Answer

Markdown literal converter (markdownToHtml) turns ChatGPT-style **bold**, # headings, fences, and pipe tables into real formatting.

Question 32

How does Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs handle: Markdown literal converter (markdownToHtml) turns ChatGPT-style **bold**, # hea…?

Accepted Answer

Markdown literal converter (markdownToHtml) turns ChatGPT-style **bold**, # headings, fences, and pipe tables into real formatting.

Question 33

What happens in approach step 7 of Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs?

Accepted Answer

Single-URL smart insert (parsePasteUrl) embeds YouTube/Vimeo videos or image nodes from bare links.

Question 34

How does Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs handle: Single-URL smart insert (parsePasteUrl) embeds YouTube/Vimeo videos or image no…?

Accepted Answer

Single-URL smart insert (parsePasteUrl) embeds YouTube/Vimeo videos or image nodes from bare links.

Question 35

What happens in approach step 8 of Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs?

Accepted Answer

FastAPI 5-stage server pipeline (linebreak, paragraph, heading, table, structure) emits Lexical JSON with per-stage telemetry — parallel to live client path.

Question 36

How does Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs handle: FastAPI 5-stage server pipeline (linebreak, paragraph, heading, table, structur…?

Accepted Answer

FastAPI 5-stage server pipeline (linebreak, paragraph, heading, table, structure) emits Lexical JSON with per-stage telemetry — parallel to live client path.

Question 37

What happens in approach step 9 of Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs?

Accepted Answer

Opt-in AI layer: 20 Next.js /api/ai routes, SSE streaming, below-selection confirm flow, ghost autocomplete with versioned abort, and sanitize_for_llm guardrails.

Question 38

How does Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs handle: Opt-in AI layer: 20 Next.js /api/ai routes, SSE streaming, below-selection conf…?

Accepted Answer

Opt-in AI layer: 20 Next.js /api/ai routes, SSE streaming, below-selection confirm flow, ghost autocomplete with versioned abort, and sanitize_for_llm guardrails.

Question 39

What challenge 1 did Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs face?

Accepted Answer

PDF copy-paste breaks paragraphs into ragged half-lines

Question 40

How was "PDF copy-paste breaks paragraphs into ragged half-lines" solved in Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs?

Accepted Answer

Line-break repair joins lines where the previous doesn't end in terminal punctuation and the next starts lowercase; spaCy sentence-boundary fallback on the server pipeline.

Question 41

What challenge 2 did Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs face?

Accepted Answer

Word/Docs paste carries kilobytes of junk HTML and fake-bold spans

Question 42

How was "Word/Docs paste carries kilobytes of junk HTML and fake-bol…" solved in Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs?

Accepted Answer

Dedicated cleanHtml sanitizer strips mso-*, , , empty spans, and normal-weight / before ProseMirror re-parses through schema.

Question 43

What challenge 3 did Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs face?

Accepted Answer

Browsers wrap plain-text copies in trivial HTML, hiding real structure

Question 44

How was "Browsers wrap plain-text copies in trivial HTML, hiding rea…" solved in Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs?

Accepted Answer

isBrowserPlainTextHtml detects trivial-tag shape and routes to parseStructuredBlocks instead of trusting noise HTML.

Question 45

What challenge 4 did Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs face?

Accepted Answer

ChatGPT output pastes markdown as literal ** and # text

Question 46

How was "ChatGPT output pastes markdown as literal ** and # text" solved in Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs?

Accepted Answer

markdownToHtml converts markdown literals into real formatting with no LLM — headings, emphasis, lists, tables, code fences.

Question 47

What challenge 5 did Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs face?

Accepted Answer

AI output could silently change the user's text

Question 48

How was "AI output could silently change the user's text" solved in Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs?

Accepted Answer

Stream generated text below the selection with mandatory confirm/cancel/undo gate — nothing replaces document text without explicit confirm.

Question 49

What challenge 6 did Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs face?

Accepted Answer

Pasted content could carry prompt-injection payloads into AI features

Question 50

How was "Pasted content could carry prompt-injection payloads into A…" solved in Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs?

Accepted Answer

sanitize_for_llm: NFKD normalization, base64 rescan, injection denylist; content wrapped as data, not instructions.

Question 51

What result 1 did Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs achieve?

Accepted Answer

A paste pipeline that turns PDF, Word/Docs, spreadsheet, markdown-literal, and structured-report content into clean nodes with zero LLM calls.

Question 52

What result 2 did Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs achieve?

Accepted Answer

Deterministic, instant, offline-capable paste cleanup locked by golden fixtures on both client (Vitest) and server (pytest).

Question 53

What result 3 did Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs achieve?

Accepted Answer

Cleanly separated opt-in AI layer: 9 rewrite modes, 20 AI endpoints, SSE streaming, ghost-text autocomplete, and construction-draft generator — none run on paste.

Question 54

What result 4 did Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs achieve?

Accepted Answer

Safety posture where AI never overwrites text silently and pasted content is never trusted as instructions.

Question 55

What result 5 did Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs achieve?

Accepted Answer

Per-feature cost observability with Chart.js dashboard.

Question 56

What did the team learn from Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs (insight 1)?

Accepted Answer

Cleanup is a structure problem, not a language problem — pure functions over clipboard strings made paste fast, free, and testable in a way an LLM never could.

Question 57

What did the team learn from Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs (insight 2)?

Accepted Answer

Determinism is a feature you can test — golden fixtures turn "it usually looks right" into a contract.

Question 58

What did the team learn from Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs (insight 3)?

Accepted Answer

Browsers lie about clipboard content — plain text arrives wrapped in trivial HTML, so disambiguation matters.

Question 59

What did the team learn from Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs (insight 4)?

Accepted Answer

Schema-validated re-parsing is free security — ProseMirror's schema drops unknown nodes without a hand-written allowlist.

Question 60

What did the team learn from Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs (insight 5)?

Accepted Answer

Isolation prevents scope creep — separating AI routes from the paste extension kept the no-LLM-in-paste rule enforceable.

Question 61

How was Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs built?

Accepted Answer

Built Smart Text Editor: TipTap paste pipeline with Office HTML sanitizer, structured-doc parser, CSV/markdown converters, and URL embeds — zero LLM in paste. Opt-in AI layer: 9 rewrite modes, SSE streaming, ghost autocomplete, confirm-before-apply.

Question 62

What was the outcome of Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs?

Accepted Answer

Instant deterministic paste cleanup locked by golden fixtures. 20 opt-in AI endpoints with preview-before-apply and prompt-injection guardrails. Judgment-driven design: structure problems don't need LLMs.

Question 63

What did Solo Builder learn from Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs?

Accepted Answer

Key learnings include Cleanup is a structure problem, not a language problem — pure functions over clipboard strings made paste fast, free, and testable in a way an LLM never could., Determinism is a feature you can test — golden fixtures turn "it usually looks right" into a contract., Browsers lie about clipboard content — plain text arrives wrapped in trivial HTML, so disambiguation matters., Schema-validated re-parsing is free security — ProseMirror's schema drops unknown nodes without a hand-written allowlist., Isolation prevents scope creep — separating AI routes from the paste extension kept the no-LLM-in-paste rule enforceable., Streaming UX needs generation-versioning — without versioned requests and abort-on-unmount, ghost text and rewrites flicker and race..

Question 64

What was Solo Builder's contribution to Smart Text Editor — A Rich Text Editor That Cleans Pasted Content With Engineering, Not LLMs?

Accepted Answer

Dhruvil Patel contributed across Built the deterministic SmartPaste pipeline — convertPastedContent, cleanHtml, parseStructuredBlocks, CSV/markdown/URL detectors, and Turndown normalization., Built the FastAPI 5-stage paste pipeline with golden-fixture pytest coverage and Lexical JSON output., Built 20 Next.js AI route handlers with SSE streaming, mode→prompt mapping, and per-feature cost tracking., Built ghost-text AiAutocomplete extension with debounce, generation versioning, Tab/Arrow accept, and Escape abort., Built below-selection confirm/cancel/undo apply flow — AI never silently overwrites document text., Built sanitize_for_llm prompt-injection defense and provider-agnostic LLMClient (Anthropic, OpenAI, Ollama, Echo)., Wrote Vitest golden fixtures for paste parsers and pytest suites for pipeline, guardrails, and construction-draft., with ownership from design through deployment.