FreshFront

Freshfront is your AI-powered creative co-founder. It unifies your data with deep research, multi-modal content generation and full marketing automation in a single workspace to help maximize output.

Comment

Inspiration

I was initially inspired to build a research tool for crypto and stock investments. As I developed the modular widget system for dynamically displaying different companies and tickers, I realized this architecture could power something much bigger—a general research tool that adapts its widgets based on topic relevancy and pre-populates them with relevant data. This sparked the vision for FreshFront: a platform that transforms how people research, create, and share digital products.

What it does

FreshFront is an AI-powered workspace that combines research, content creation, and marketing automation into one seamless platform. Users can conduct deep research with enhanced web search and red teaming capabilities, save reports to a living dashboard, and manage multi-modal assets. The platform supports over 20 asset formats including blogs, images, videos, PDFs, tables, 3D worlds, products, websites, and lead forms. With built-in email and cross-platform social media scheduling, users can integrate assets into campaigns with one click. An intelligent assistant with chat and real-time voice interfaces provides access to specialized agents (research, content creation, marketing, browser automation, project management) that work together to execute complex tasks. The Project Context "oracle" acts as a safeguard, reviewing all activity and intelligently managing what information gets passed to the assistant, ensuring efficiency using state-of-the-art flash models.

How we built it

FreshFront is built on a modular architecture with multiple specialized agents coordinated by a central assistant. I implemented intent recognition using Gemini Flash with thinking to ensure accurate tool selection before execution. The platform features a comprehensive asset management system with file search/RAG capabilities, a visual node map for note-taking, and source-stacking with reverification systems that replace outdated data automatically. For the voice interface, I packaged complex chat workflows into lighter versions optimized for real-time interaction while maintaining transparency by displaying all processes to users. The backend includes robust cron job systems for reliability, handling large file sizes and queue management. Context window management uses intelligent chunking where specialized agents cherry-pick relevant context chunks rather than sending everything for each request.

Challenges we ran into

Creating a seamless real-time AI voice and chat assistant with extensive agent tool calls was a major challenge. I had to recognize the critical importance of intent recognition and leverage Gemini Flash with thinking to avoid incorrect tool usage and ensure decisions align with user goals. Packaging heavy chat workflows into lighter versions for voice mode took weeks of trial and error before achieving a system that's both comprehensive and fast. Optimizing cron job reliability while managing large file sizes and queue jams required rigorous testing across different scenarios to confirm repeatable expected results. Managing context windows became a complex puzzle—preventing API failures from oversized contexts while ensuring relevant information gets through.

Accomplishments that we're proud of

I created the ideal tool I needed that didn't exist until I built it. FreshFront is not only more feature-rich than NotebookLM but also more fun and intuitive to use—I no longer have the desire to go back. The platform seamlessly enables users to go from research to product creation: they can ask the assistant to review project data, scan social media for trends, create products with Stripe integration, browse for inspiration, build storefronts, generate content and ads, and promote across social media and email—all in one go. The color-coded project themes and context-backing keep everything organized under one cohesive ethos, making users high-level decision makers while AI handles execution.

What we learned

Using flash models with thinking for intent detection is crucial—if the agent misunderstands user intent initially, it wastes tokens on a cascade of wrong decisions. Flash models with thinking deliver accurate results quickly. Context window management is an ongoing puzzle requiring creative solutions like chunking systems where agents cherry-pick relevant context rather than overwhelming the assistant with everything. Transparency matters: displaying all processes to users retains their attention and builds trust, even when complex workflows run in the background.

What's next for FreshFront

The future looks bright with emerging technologies like Genie 3. If an API becomes available, we could upgrade from Gaussian splatting to truly interactive 3D worlds—imagine researching "The Future of Aviation" and actually driving a generated flying car based on your stacked sources, or exploring "Climate in the Pre-Historic Age" by walking through generated environments. As AI models improve, we'll continuously update the platform so users can transform their data into high-quality apps, interactive courses, YouTube videos, podcasts, and more. This modular system's strong foundation is designed to support infinite innovations to come.

Share this project:

Updates