Promptly

Inspiration

I’ll never forget when my dad forwarded a “miracle cure” to our family chat. He meant well—but it was misinformation. I kept seeing the pattern everywhere: conspiracy links, manipulative headlines, and good conversations derailed by logical fallacies.
The problem wasn’t carelessness—it was effort. Analyzing content, crafting thoughtful replies, and verifying claims meant opening tabs, drafting responses, cross-referencing sources, and checking tone. By the time you’re done, you’ve scrolled past ten more posts.

We needed something different: an intelligent companion that could summarize, reply, detect manipulation, and verify claims—all without leaving the page. That’s how Promptly was born.

What it does

Highlight any text; a floating button appears. Click to open a persistent side panel with:

🧠 Smart Replies — Contextually perfect responses (supportive, professional, questioning, counter-argument), plus custom tones (“make it funny,” “be more technical”).
📝 Summarization — Condenses articles, research papers, and threads into concise key points in seconds.
🛡️ ArgumentArmor (unique) — Detects 15 fallacies (ad hominem, strawman, false dichotomy, slippery slope, circular reasoning, etc.), explains why they’re problematic, and rates arguments on evidence quality, logical consistency, and claim support.
😊 Sentiment Analysis — Reveals emotional undertones (angry, manipulative, neutral) to guide engagement.
🛡️ Toxicity Shield — Flags harmful/aggressive language to avoid provocation traps.
⚖️ Debate Mode — Generates balanced pro/con perspectives that steelman both sides.
🌍 Echo Chamber Detector — Presents credible opposing views with citations to escape filter bubbles.
✓ Fact-Checking — Verifies claims via Google Fact Check Tools API (100+ pro fact-checkers), Wikipedia, and authoritative domains; requires multi-source consensus with transparent, clickable citations and quality notes.

Runs locally with Chrome’s Built-in AI (Gemini Nano) when available, with intelligent cloud fallback as needed.

How we built it

Platform & UI
Chrome Manifest V3 extension with a persistent side panel (not a transient popup). Flow: highlight → floating button → panel. Collapsible sections, smooth CSS transitions, modern gradient design.
Intelligence Layer (Hybrid AI)
Primary: Chrome Built-in AI (Prompt API + Summarizer API on Gemini Nano) for privacy + speed.
Fallback: Gemini 2.0 Flash (cloud) with automatic availability detection and offline-aware behavior.
Smart Reply Engine
Parses selected text for intent (question/statement/argument/opinion), chooses tone, and generates multiple response variants; supports custom style directives.
ArgumentArmor
Specialized analyzer parsing text structure and reasoning patterns; detects 15 fallacies and scores multi-dimensional argument quality; targets <3s latency.
Sentiment & Toxicity
Calibrated prompts distinguish passionate vs. toxic language and strong opinion vs. manipulation; tuned via real-world testing.
Debate Mode & Echo Chamber
Prompts explicitly steelman opposing views (no strawmen) and surface credible counter-evidence.
Fact-Checking Pipeline
Three parallel streams (Fact Check API, Wikipedia, authoritative domains), cross-referenced with consensus rules and transparent quality ratings.
Why side panel (not popup)
Analysis and verification happen alongside browsing; minimal yet powerful UI with collapsible settings and smooth transitions.

Challenges we ran into

Browser function export bug — Dev worked; build returned “undefined” (exported for Node via module.exports but not exposed to window). Fixed across four files with dual-export pattern.
Context in Smart Replies — Early outputs were generic; added intent detection (question vs. statement), sarcasm cues, domain context, and tone matching.
ArgumentArmor nuance — Differentiating strawman vs. legitimate rebuttals and false dichotomies vs. valid binaries required deep prompt engineering and debate-set testing.
UI complexity — Too many buttons overwhelmed users; solved with grouping (primary vs. advanced), progressive disclosure, and clear visual hierarchy.
Side panel lifecycle — Persisting state, handling open/close, and staying non-intrusive needed multiple iterations.
Performance — Targets: summarization <2s, replies <3s, fallacy detection near-instant; achieved via prompt optimization, caching, and smart on-device vs. cloud selection.

Accomplishments we’re proud of

First integrated browser fallacy detector that’s accurate, explanatory, and educational.
Context-aware Smart Replies that feel human and adapt to custom tones.
Instant, privacy-preserving summaries via Chrome’s on-device Summarizer.
Balanced debates that steelman opposing views.
Production-quality fact-checking with multi-source consensus and transparent sourcing (no AI “guessing”).
Gorgeous, minimal UI in a persistent side panel that feels native to Chrome.
Hybrid AI architecture prioritizing privacy (on-device first) with robust cloud fallback.
It works right now—any page, any text, nearly instantly.

What we learned

Building for Built-in AI is different — Availability checks, offline handling, and on-device constraints demand true progressive enhancement.
Context is the difference between generic and useful AI—understand why the user selected text and what they want to do next.
UX > feature count — Nine fast, intuitive features beat twenty slow ones.
Fallacy work teaches real logic — ArgumentArmor sharpened understanding of reasoning and critical thinking.
Performance is a feature — Sub-3-second responses make features feel “magical.”
Personal growth — Turning “I wish this existed” into “here, I built it.”

What’s next for Promptly

Conversation threading — Analyze full comment chains, track evolving arguments, and pinpoint where discourse derails.
Reply suggestions with evidence — Auto-attach citations and fact-checks to responses.
Tone coaching — Real-time feedback while typing (“soften this,” “acknowledge before countering”).
Education integration — Classroom modules for logic/media literacy; analyze historical speeches and research.
Custom profiles — Professional, skeptical, empathetic modes with saved preferences.
Multilingual expansion — Replies, summaries, and fallacy detection across dozens of languages.
Browser-native integration — Make ArgumentArmor, replies, and summaries core features (e.g., right-click → “Analyze with Chrome AI”).

Vision: A web where everyone has instant, trustworthy analysis—making discourse more thoughtful, informed, and constructive.

Elevator Pitch (≤200 chars)

“Instant AI analysis: smart replies, summaries, fallacy detection, sentiment & toxicity checks, and multi-source fact-checking—without leaving the page.”