“Can’t recall the name, but there was a scene where a woman walks alone in the rain, and everything is blue...”
Let AI do the guessing.
SceneSeeker AI is an AI assistant that identifies movies/TV shows based on vague scene descriptions using RAG (Retrieval-Augmented Generation) and LLMs. It bridges the gap between fuzzy memory and accurate title recognition using smart retrieval and generative intelligence.
People often remember just a scene, not the title:
“There’s a train fight and the guy wears red and black…”
Search engines fail at such queries. SceneSeeker AI interprets this ambiguity and retrieves the most likely movie/show matches using semantic search + LLM.
- Top 3–5 results containing:
- 🎬 Title
- ✅ Confidence score
- 📝 Short synopsis
- 🖼️ Optional thumbnail
-
Retriever
- Converts the user query into embeddings
- Searches a vector DB (e.g. FAISS, Chroma) for similar scene summaries
-
Generator
- Takes the top matches as context
- Uses an LLM (e.g. GPT-4, Mistral, Claude) to rank and summarize potential answers
-
Post-Processor
- Sorts results by relevance score
- Returns clean, structured output
| Layer | Tool/Tech |
|---|---|
| Frontend | React / Next.js (optional UI) |
| Backend | FastAPI / Flask |
| Embeddings | OpenAI / HuggingFace / Cohere |
| Vector DB | FAISS / Chroma |
| LLM | GPT-4 / Claude / Mixtral |
| Storage | JSON / MongoDB (optional) |
“The movie where a guy writes letters to his past self using a mailbox”
“The animated film with emotions as characters in a girl’s head”
“A detective with amnesia wakes up with tattoos all over his body”
User Input:
“I remember this movie where a woman gets letters from a man living in a different year. They meet through a mailbox, I think?”
AI Output:
{
"guess": "The Lake House",
"confidence": "High",
"reason": "The Lake House features time-travel letters exchanged via a mailbox and stars Sandra Bullock.",
"related": ["Your Name", "The Time Traveler's Wife", "About Time"]
}