/* This project is an argument summarizer that leverages AI to analyze and extract key arguments from a given text. Goals: 1. Provide a user-friendly interface for inputting text 2. Process the input using a large language model (LLama3 via Groq) 3. Extract and structure key arguments, explanations, and relevant quotes 4. Present the summarized arguments in a clear, organized format The main pipeline: 1. User inputs text through a web interface 2. The input is sent to an AI model for processing 3. The AI extracts and structures the arguments 4. The results are validated against a predefined schema 5. The structured arguments are displayed to the user This tool aims to help users quickly understand the main points and supporting evidence in complex texts or discussions, making it valuable for research, debate preparation, or general comprehension of argumentative content. */ import { Hono } from "https://deno.land/x/hono@v3.11.7/mod.ts"; // Inlined content from ./middleware.js import { z } from 'npm:zod'; import { zerialize } from "npm:zodex"; export const pipeline_url = `https://yawnxyz-executeproxy.web.val.run/execute` // const pipeline_url = `http://localhost:9999/execute`; export const useCache = true; export const saveCache = true; // const model = "llama-3.1-8b-instant" // const provider = "groq" // keeps running out, but better than mini // const model = "models/gemini-1.5-flash-latest" // const model = "models/gemini-1.5-pro-latest" // const provider = "google" // export const model = "gpt-4o-mini" // export const model = "gpt-4o" // export const provider = "openai" export const summaryModel = "gpt-4o-mini" export const summaryProvider = "openai" // export const smartModel = "models/gemini-1.5-pro-latest" // export const smartProvider = "google" export const smartModel = "gpt-4o-mini" export const smartProvider = "openai" export const cheapModel = "gpt-4o-mini" export const cheapProvider = "openai" const argschema = z.object({ arguments: z.array(z.object({ argument: z.string(), explanation: z.string(), quote: z.string() })) }); const zargschema = zerialize(argschema); const critschema = z.object({ critiques: z.array(z.object({ critique: z.string(), explanation: z.string(), quote: z.string() })) }); const zcritschema = zerialize(critschema); export const execute2 = async (c) => { const body = await c.req.json(); const input = body.input; // console.log('[execute] input:', input); const pipeline = [ { "name": "concurrent", "steps": [ { "name": "ai", "save": "arguments", "settings": { "system": "You are an AI assistant tasked with analyzing text and generating a summary of thought-through arguments. Follow these steps:\\n1.) Analyze the input text and identify 5 key arguments or points.\\n2.) For each argument:\\n a. Clearly state the argument or point\\n b. Provide a detailed explanation or analysis\\n c. Include a relevant, direct quote from the text to support the argument\\n3.) Output the result as a JSON object with 'argument', 'explanation', and 'quote' keys for each point. Don't include any other text. ONLY OUTPUT JSON. Do not introduce yourself. Important: If the article content seems to be missing or locked behind a paywall, respond to say that the content is missing.", "prompt": input, "model": smartModel, "provider": smartProvider, "structuredOutputs": true, "outputType": "object", "schemaName": "argumentAnalysis", "schemaDescription": "Analysis of key arguments from the input text", "zschema": zargschema, } }, { "name": "ai", "save": "critiques", "settings": { "system": "You are an AI assistant tasked with constructively critiquing the arguments of this text and generating a summary of your critiques. Please critique from a meta, academic level, like a peer reviewer would do. Pick apart the arguments. If a research paper, did the authors follow through? Is the data thorough? Are the arguments cohesive? Did the article or paper support their arguments?. Follow these steps:\\n1.) Evaluate the input text and identify 5 key critiques or points.\\n2.) For each critique:\\n a. Clearly state the critique or point\\n b. Provide a detailed explanation or analysis of your critique\\n c. Include a relevant, direct quote from the text to support the critique or evaluation\\n3.) Output the result as a JSON object with 'critique', 'evaluation', and 'quote' keys for each point. Don't include any other text. ONLY OUTPUT JSON. Do not introduce yourself. Important: If the article content seems to be missing or locked behind a paywall, respond to say that the content is missing.", "prompt": input, "model": smartModel, "provider": smartProvider, "structuredOutputs": true, "outputType": "object", "schemaName": "critiqueAnalysis", "schemaDescription": "Analysis of key critiques from the input text", "zschema": zcritschema, } }, ] }, { "name": "concurrent", "steps": [ { "name": "ai", "save": "summary", "settings": { "system": "Be concise however makes sense. Just output the bullet points in markdown, no intro, no explanation, no commentary. Don't introduce the response just write it.", "prompt": "Please combine the analysis and critiques in to a concise, 2-4 sentence paragraph that contains an overarching balanced summary, using each argument's analysis. Use simple english and don't use any fancy words. At the end of the summary add 3-5 bullet points that are the most important points to take away from the analysis: {arguments} and critique's analysis: {critiques}. Just output the balanced summary, no intro, no explanation, no commentary. Don't introduce the summary just write it.", "model": smartModel, "provider": smartProvider, } }, { "name": "ai", "save": "critiquesummary", "settings": { "system": "Be concise however makes sense. Just output the bullet points in markdown, no intro, no explanation, no commentary. Don't introduce the response just write it.", "prompt": "Please write 2-3 sentences that contains an overarchig summary of the arguments. Use short sentences and simple english. Then create a list of 2-3 of the most important key bullet points, in markdown, for the following critique summary: {critiques}. Just output the bullet points in markdown, no intro, no explanation, no commentary. Don't introduce the response just write it.", "model": smartModel, "provider": smartProvider, } }, { "name": "ai", "save": "title", "settings": { "prompt": "Please create a short 3-8 word, nuanced and balanced title for the following summary:

{summary}

. Just output the title, no intro, no explanation, no commentary. Don't introduce the response just write it.", "model": smartModel, "provider": smartProvider, } }, ] }, { "name": "data", "settings": { "data": { "title": "{title}", "arguments": "{arguments}", "critiques": "{critiques}", "summary": "{summary}", "summaryBullets": "{summaryBullets}", "critiquesummary": "{critiquesummary}", } } } ] const response = await fetch(pipeline_url, { method: "POST", headers: { "Content-Type": "application/json" }, body: JSON.stringify({ verbose: true, useCache: useCache, saveCache: saveCache, pipeline }) }); const result = await response.json(); // console.log('[execute] result:', result.result.data.result.arguments); const args = JSON.parse(result.result.data.result.arguments).object.arguments; const critiques = JSON.parse(result.result.data.result.critiques).object.critiques; console.log('[execute] result args:', args); return c.json({ arguments: args, critiques: critiques }); }; // Add new route for fetching source content export const fetchSource = async (c) => { const body = await c.req.json(); const url = body.url; if (!url) { return c.json({ error: "URL is required" }, 400); } const pipeline = [{ "name": "getContent", "settings": { "url": url, "optionStr": "content,doi,citation", "crawlMode": "jina", "outputMode": "markdown" } }]; try { const response = await fetch(pipeline_url, { method: "POST", headers: { "Content-Type": "application/json" }, body: JSON.stringify({ useCache: useCache, saveCache: saveCache, pipeline }) }); if (!response.ok) { throw new Error(`Failed to fetch content from ${url}`); } const result = await response.json(); return c.json({ content: result.content || result.text || result }); } catch (error) { console.error(`Error fetching source ${url}:`, error); return c.json({ error: error.message }, 500); } }; // Inlined content from ./styles.js export const tailwindConfig = { theme: { extend: { colors: { 'background': '#F5F4F4', 'page-bg': '#F5F4F4', 'app-bg': '#FDFBF7', 'app-bg-alt': '#FFFFFF', 'app-accent': '#E2DFDF', 'primary': '#0E141B', 'accent': '#FDFAF4', 'secondary': '#D9D5D4', 'text': '#100C04', 'button': '#39516A', 'button-hover': '#0063CC', } } } }; export const tailwindStyles = ` @tailwind base; @tailwind components; @tailwind utilities; html { @apply bg-background; } .AppBody { @apply bg-page-bg text-text py-4; } .AppLogo { @apply max-w-2xl mx-2 md:mx-auto pb-2; } .AppLogo img { @apply w-8 h-8; } .AppContainer { @apply container mx-auto; } .App { @apply max-w-2xl mx-2 md:mx-auto p-4 bg-app-bg border border-app-accent rounded-md shadow-sm; } .AppFooter { @apply mt-4 text-center text-sm; } .AppFooter a { @apply text-primary hover:text-button-hover underline; } h1 { @apply text-3xl font-semibold mb-2 text-primary; } p { @apply pb-4; } a { @apply text-primary underline hover:text-button-hover; } textarea { @apply w-full mb-2 py-2 px-4 bg-app-bg-alt border border-app-accent rounded-md focus:outline-none focus:ring-1 focus:ring-secondary; } button { @apply bg-button text-white px-4 py-2 rounded-md hover:bg-button-hover transition duration-300; } button:disabled { @apply opacity-50 cursor-not-allowed; } #result { @apply mt-4 p-4 bg-accent border border-app-accent rounded-md text-sm whitespace-normal; } .history-item { @apply flex flex-row justify-between items-center p-2 bg-accent hover:bg-app-bg-alt border border-app-accent rounded-md cursor-pointer hover:bg-app-bg-alt transition-colors; } .delete-button { @apply self-end bg-button text-white px-2 py-1 rounded text-sm hover:bg-button-hover transition-colors; } button.accordion { @apply bg-app-bg text-primary px-0 py-0 rounded-md transition-colors hover:text-button-hover; } ul li { @apply list-disc list-outside ml-5; } .input-textarea { @apply min-h-[60px] max-h-[20svh]; } .input-text { @apply mt-2 p-2 bg-app-bg-alt border border-app-accent rounded-md whitespace-pre-wrap overflow-scroll max-h-[300px] text-xs; } `; const app = new Hono(); // app.post("/execute", execute); app.post("/execute2", execute2); app.post("/fetch-source", fetchSource); const module = ` `; const html = ` Breakdown

Breakdown

This tool breaks down any text into arguments and critiques, and attempts to support with citations from the text. This works best with short texts like abstracts, journal articles, news articles, and press releases.

Try using these examples below:

https://www.nature.com/articles/srep26717
https://arstechnica.com/health/2024/11/as-hospitals-struggle-with-iv-fluid-shortage-nc-plant-restarts-production

Inspiration: rohanpaul_ai | ${smartProvider}/${smartModel}

Show Input Text

This was the text that was fed into the AI model.

A labspace.ai experiment

${module} `; app.get("/", (c) => c.html(html)); export default (typeof Deno !== "undefined" && Deno.env.get("valtown")) ? app.fetch : app;

breakdown