Please sign up or log in to continue.

My ET: Your Personal News Podcast Studio

AI-powered news assistant that translates complex articles into any language , creates podcasts, verifies facts to stop fake news, Making knowledge accessible to everyone.

Comment

My ET: Your Personal News Podcast Studio

Inspiration

We noticed three big problems regular people face when trying to consume and share information:

Too Much to Read, Not Enough Time to Listen: People want to stay informed, but reading long articles is exhausting. Commuters, students, and busy professionals would much rather listen — but turning any article or topic into a real podcast takes hours of scripting, recording, and editing.

Language Barriers: Millions of people cannot access quality content because it is only available in one language. This locks out a huge population from staying informed and engaged in conversations that matter to them.

Misinformation Spreads Fast: A false message goes viral, thousands share it, and by the time the truth catches up — the damage is done. People need a fast, reliable way to verify what is real before they hit share.

We asked ourselves: What if turning any article, topic, or idea into a podcast was as easy as pressing one button?

What it does

My ET transforms passive content consumption into an active podcast creation studio. Think of it as having a professional scriptwriter, voice director, translator, and fact-checker in your pocket — ready to turn anything you are reading into something worth listening to.

Podcast Studio

One click turns any article, topic, or idea into a fully produced podcast episode:

  • Choose your format: 2-person conversational, solo narration, interview-style Q&A, or short daily brief
  • AI generates natural, human-sounding scripts — not robotic summaries
  • Multi-agent pipeline: content analysis, script generation, voice direction, audio synthesis
  • Works on any content: news articles, blog posts, your own notes, or just a topic you type in

Smart Content Search

Find what you want to turn into a podcast — fast:

  • Search any topic and get AI-curated source material ready to convert
  • Pull from live news, blogs, and reference content simultaneously
  • Example: Search "climate policy 2025" and get a ready-to-record 5-minute episode brief in seconds

Multilingual Audio

One-click podcast generation across 10+ languages — not literal translation:

  • Powered by Gemini and language-specific models for cultural and linguistic nuance
  • Scripts adapt idioms and metaphors to feel native, not translated
  • Makes quality audio content accessible to audiences worldwide regardless of their primary language

Truth Engine

Upload an article, screenshot, or claim and AI verifies it before you turn it into content:

  • Vision AI analyzes images and text
  • Cross-references against authoritative sources and official records
  • Returns a verdict — true, false, or misleading — with clickable citations
  • Ensures every podcast you create is built on verified information

AI Studio

Nine content tools available for every topic:

  1. Podcast Script: Full 2-person conversational script, ready to record
  2. Solo Brief: Tight single-host narration for commuters, under 5 minutes
  3. Interview Format: Host and guest Q&A structure with suggested follow-ups
  4. Episode Arc: Intro, story, insight, and outro — auto-structured
  5. Context Cards: Hover over jargon in your script and get instant plain-language definitions
  6. Contrarian Angle: Devil's advocate perspective to make episodes more engaging
  7. Multilingual Script: Full script rewritten in your chosen language
  8. Video Brief: 60-second short-form script with visual cues for reels and shorts
  9. Fact Score: Credibility rating for your source material, validated before you record

Performance

  • SHA-256 Caching: Repeat queries drop from 8 seconds to under 40ms
  • Cached in Neon Postgres using content-hash keys
  • Fast enough to feel instant, even on low-bandwidth connections

How we built it

Tech Stack

  • Frontend: Next.js 16 (App Router), Tailwind CSS v4, Shadcn UI, Framer Motion
  • AI Models: Eleven Labs for language-specific TTS models for multilingual audio, Gemini for script generation and synthesis, Gemini 2.5 Pro for vision and misinformation detection,
  • Search and Verification: Tavily API for content aggregation and fact-checking
  • Database: Neon Postgres with Drizzle ORM for user profiles, script library, and caching
  • Authentication: Clerk for secure user personas

Multi-Agent Architecture

  1. Router Agent: Understands intent — topic search, script generation, or fact-check
  2. Content Agents: Parallel fetch and synthesis from multiple sources
  3. Script Agent: Generates format-specific scripts — conversational, solo, interview
  4. Persona Agent: Adapts tone and complexity based on user profile
  5. Voice Agent: Directs TTS output with pacing, emphasis, and language cues

Key Technical Innovations

  • Format-aware scripting: Different prompting strategies per podcast format produce genuinely distinct output styles, not just reformatted text
  • Deterministic data guardrails: Factual claims in scripts are always grounded in fetched source material, never AI-generated from memory
  • Content-hash caching: SHA-256 hash of query plus tool ID enables instant cache retrieval
  • Hybrid translation: Semantic accuracy from LLMs combined with linguistic nuance from language-specific models

Challenges we ran into

AI Hallucination in Scripts: Early versions invented facts and quotes. We fixed this by grounding every script in fetched source content — the AI shapes language, not facts.

Translation Quality: Literal translations of idioms and culturally specific phrases produced awkward, unnatural scripts. We built a hybrid pipeline that replaces expressions with locally meaningful equivalents.

Performance Bottleneck: Initial generation took 8 to 12 seconds. SHA-256 caching in Neon Postgres brought repeat queries to under 40ms.

Misinformation Detection Accuracy: Vision models struggled with low-quality screenshots. We improved this with enhanced preprocessing and cross-referencing against multiple authoritative sources.

Format Consistency: Keeping conversational scripts sounding natural — not like summarized bullet points read aloud — required significant prompt engineering and format-specific evaluation.

Accomplishments we are proud of

Zero hallucination guarantee: Scripts are grounded in real source material, not AI imagination.

Genuine multilingual support: Not translation — cultural adaptation that makes content feel native in any language.

40ms response time on cached queries: Enterprise-grade performance on consumer hardware.

Misinformation shield: Fact verification before content creation prevents false information from being amplified.

What we learned

AI is only as good as its sources: The most capable model produces unreliable output if it is not grounded in real data. Source fetching is not optional for a content tool — it is the foundation.

Natural language is hard: There is a significant difference between a script that contains the right information and a script that actually sounds like two people talking. Format-specific prompt engineering matters enormously.

Caching is a UX feature: Users do not care about infrastructure costs — they care about speed. Sub-100ms response times change how people feel about a product.

Misinformation is a technical problem: Better content alone does not stop false information from spreading. You need multi-source verification, vision AI, and authoritative references working together.

Personalization drives retention: Generic outputs are forgettable. Users stay when the platform understands their topic, their format, and their audience.

What is next

  • Expand to 50+ languages with voice output, not just script translation
  • API for developers: Let media companies and content tools build on our podcast generation layer
  • Autonomous podcast agents: Monitor a topic continuously and generate new episodes when something significant happens
  • Enterprise tier: White-label podcast studio for newsrooms, educators, and corporate communications teams

Business model

  • Free tier: 3 podcast scripts per day, standard formats
  • Creator plan at $9 per month: Unlimited scripts, multilingual output, audio export
  • Studio plan at $29 per month: Voice customization, RSS publishing, collaborative editing
  • Enterprise: Custom pricing for newsrooms and corporate teams

My ET is not just a content tool — it is a movement to make every person a broadcaster. Whatever you are reading, whatever you are thinking, whatever story you want to tell — we make it listenable.

Built With

Share this project:

Updates