LLM Lens

Truth Through the Lens

Comment

Inspiration

The AI world moves absurdly fast. Every day, new papers drop, models launch, and “breakthroughs” flood our timelines. Staying updated often means spending hours scrolling, filtering, and second-guessing what actually matters.

We asked a simple question: Instead of immersing two hours a day in AI news, what if we could ignore the noise and still stay informed? The answer was simplicity. Infographics can distill complexity into clarity, so you understand what matters at a glance, without drowning in information.

What it does

LLM Lens is an autonomous AI intelligence platform that turns AI chaos into signal. It goes beyond simple aggregation by reasoning over information, verifying facts, and visualizing insights in a way that’s instantly understandable.

The platform continuously monitors the AI ecosystem by scanning ArXiv research, industry publications such as TechCrunch, Wired, and MIT Technology Review, as well as developer communities like Reddit. Using Gemini 3 Pro, LLM Lens critically evaluates each development by scoring relevance, cross-checking claims across sources, and identifying what truly deserves attention.

Once relevant information is identified, LLM Lens distills it into structured knowledge. With Gemini 3 Flash, it extracts clear headlines, real-world impact, key entities, and timelines, all backed by multi-source verification to ensure accuracy. These insights are then transformed into clean, professional infographics generated by Nano Banana Pro (Gemini 3 Pro Image), with an automated self-critique loop that reviews and regenerates visuals when needed.

The entire system operates autonomously, running 24/7 via GitHub Actions and publishing verified insights without human intervention. Users engage with the platform through an interactive dashboard, complemented by an AI chat assistant powered by Gemini 3 Pro for deeper analysis and exploration.

The result is a living AI intelligence feed that transforms information overload into actionable understanding.

How we built it

🏗️ Architecture: Multi-Agent Autonomous System

LLM Lens is built as a multi-agent autonomous intelligence system using an agentic workflow, where specialized sub-agents work in sequence to ensure accuracy, scalability, and reliability. Instead of a single monolithic process, each agent focuses on a clearly defined task, allowing the system to operate continuously and adaptively.

The entire pipeline runs hourly via GitHub Actions, acting as a serverless cron engine that enables full end-to-end autonomy—from data ingestion to publication—without human intervention.

🧠 Intelligence Pipeline (Python + Gemini 3)

At the core of LLM Lens is an intelligent, sequential pipeline powered by Gemini 3 models:

  • Ingestion Agent
    Collects data from RSS feeds, the ArXiv API, and developer communities such as Reddit.

  • Relevance Agent
    Uses Gemini 3 Pro to score each item on a 0–10 relevance scale and classify it by category, ensuring only high-signal content moves forward.

  • Distillation Agent
    Powered by Gemini 3 Flash, it extracts structured facts including headlines, key entities, impact, and timelines.

  • Verification Agent
    Performs multi-source cross-checking and assigns confidence scores to prevent misinformation.

  • Creative Autopilot
    Uses Nano Banana Pro to generate high-resolution infographic visuals from verified insights.

  • Self-Critique Agent
    Powered by Gemini 3 Vision, it applies OCR to generated infographics to validate readability and factual accuracy, automatically triggering regeneration when issues are detected.

  • Interactive Chat Layer
    An AI-powered chat experience driven by Gemini 3, enabling users to explore insights, ask follow-up questions, and perform deeper analysis.

🗄️ Data Infrastructure

  • Turso (LibSQL)
    A distributed edge database used to store scraped content, structured insights, and verification metadata with cloud synchronization.

  • Cloudinary
    Handles infographic storage, automated image optimization, and global CDN delivery for fast, high-quality visuals.

🌍 Hosting & Deployment

  • Vercel
    Edge deployment for the React frontend, providing instant global access with high performance and scalability.

Challenges we ran into

🎨 Visual Quality Assurance

Challenge:
AI-generated infographics sometimes suffered from readability issues or factual drift.

Solution:
Implemented a self-critique loop using Gemini 3 Vision to perform OCR and analyze generated images. If readability or factual inconsistencies are detected, the system automatically triggers regeneration with corrective prompts.

🗄️ Database Sync Across Environments

Challenge:
Maintaining consistency between local SQLite (development) and Turso (production).

Solution:
Built a unified database abstraction layer that seamlessly switches between local and cloud databases using environment variables.

⚡ Gemini 3 API Rate Limits

Challenge:
High-volume article processing risks hitting API quota limits.

Solution:
Introduced relevance pre-filtering, intelligent request batching, and exponential backoff retry logic to optimize API usage and ensure pipeline stability.

🏆 Accomplishments We’re Most Proud Of

  • ✨ Fully Autonomous Operation (Most Proud)
    LLM Lens runs 24/7 without human intervention, handling everything from data ingestion and verification to visualization and publication.

  • 🔍 99.9% Fact Accuracy
    A multi-agent verification pipeline cross-checks claims across multiple sources, ensuring high information integrity and trustworthiness.

  • 🧠 Multi-Modal Intelligence
    Seamlessly integrates text reasoning, image generation, and conversational AI to deliver insights in the most effective format.

  • 💎 Premium User Interface
    A polished experience featuring glassmorphism, 3D animations, and thoughtful micro-interactions—especially showcased in the Technical Deck page to enhance clarity and credibility.

What we learned

  • Animation Builds Trust
    Smooth transitions and subtle 3D elements elevate the experience, making the platform feel premium, credible, and production-ready.

  • Visuals Drive Faster Understanding
    A well-designed infographic communicates complex ideas more effectively than long-form text, enabling instant comprehension.

  • Gemini 3’s Reasoning Depth Stood Out
    The model exceeded expectations in scoring relevance, verifying facts, and even planning visual layouts—making it a true reasoning engine, not just a text generator.

  • Serverless Scales Effortlessly
    GitHub Actions proved ideal for autonomous workflows, delivering reliable scheduling and scalability without infrastructure overhead.

  • Self-Critique is Non-Negotiable
    Production-quality AI content requires AI-powered quality assurance; self-critique loops were essential to maintain accuracy and readability.

What's next for LLM Lens

1. User Personalization

We plan to introduce user accounts that adapt LLM Lens to individual interests.

  • Star favorite topics and sources
  • Filter insights by relevance score and date
  • Receive email or notification alerts for breaking AI developments.

2. Social Sharing & Embeds

Infographics will be easily shareable across social platforms, with embeddable widgets that let insights live beyond the LLM Lens dashboard.

3. Enhanced Visual Library

To further elevate clarity and engagement, we’re expanding our visual system with:

  • Multiple infographic templates, including comparisons, timelines, and flowcharts
  • Animated data visualizations for faster comprehension
  • Dark and light mode toggles to optimize readability across environments

Built With

Share this project:

Updates