Inspiration
We’ve all woken up from dreams that left us feeling confused, inspired, or haunted. Yet, these fleeting stories often vanish before we can make sense of them. We wanted to build something that preserves those fragile nighttime thoughts — and helps users find meaning in them. The idea of combining journaling with AI interpretations and story generation came from a desire to blend human introspection with machine creativity.
What it does
DreamDecoder is a dream journaling platform that allows users to: Record daily journal entries or specific dreams Use AI to interpret dreams — analyzing symbols, emotions, and hidden meanings Generate immersive stories from dream content View statistics and trends about their dreams, moods, and themes
Export dream logs or stories into printable PDFs
It's more than a diary — it's a creative mental health companion.
How we built it
Frontend: Built using React and Vite for fast, modular UI. Integrated Google Fonts and a dreamlike theme using custom CSS variables. Backend: FastAPI handles all API endpoints for journal, dreams, interpretation, story generation, and more. AI Integration: Interpretation and story generation routes powered by Cohere and LLaMA-based models, via a local inference server or APIs. Dev: Replit used as a collaborative coding environment. Data: Stored in a local ChromaDB, accessible via RESTful APIs.
Challenges we ran into Configuring local LLMs (LLaMA) to respond reliably without overloading CPU.( My laptop didn't have a GPU) Designing UI/UX that balanced whimsy with clarity, especially when visualizing dream data. Adapting AI outputs (interpretation, story) into UI components while keeping latency low. Time crunch , Had other commitments and couldn't accomplish the Story visualization.
Accomplishments that we're proud of
Built a full-stack journaling application with multiple intelligent features in under 48 hours Designed an intuitive, mobile-first interface with dreamy visual theming Implemented local AI interpretation that actually made people say "wow" Created a seamless flow from dream entry ➝ interpretation ➝ story
What we learned
How to run and interact with local LLMs like LLaMA Using prompt engineering to extract emotion, symbols, and themes from freeform text Building scalable FastAPI routes and connecting them to React Importance of caching, loading states, and AI fallbacks in real-time systems That dreams — and code — are both best understood in layers
🚀 What's next for DreamDecoder Add user authentication and profiles Add StoryBoard , create visualizations of story Enable social sharing or anonymous dream community boards Add daily prompts for journaling and lucid dream training Enhance interpretation accuracy with fine-tuned models ( while this was coded I couldn't actually use it due to hardware constraints ) Use embeddings to cluster and analyze common themes across dreams
Log in or sign up for Devpost to join the conversation.