Find any movie from a few seconds of video clip

Paste a TikTok, Instagram, or YouTube clip and Ovelo names the exact film or show in seconds, using what is said, who is on screen, and a live web search.

Role

Solo Engineer, Client and Server

Organisation

Self-directed

Industry

Consumer · Media · Applied AI

Focus

Mobile · AI · Full-stack

/ Problem

People constantly see a clip and want to know what it is from, but any single clue is unreliable: dialogue is ambiguous, one face appears in many titles, and short clips carry little to go on. The goal was a fast, accurate answer from just a few seconds of video.

/ Solution

The server probes the clip, then transcribes the audio and recognises the on-screen faces in parallel. A reasoning model resolves those clues into one title through a targeted web search, then enriches it with poster, cast, and trailer. A WebSocket streams each stage back to the phone, and the pipeline is hardened with rate limiting, caching, and request de-duplication.

Version 1 built a vector database of transcript embeddings, but I realised I would be paying to index effectively the entire world of film and TV. The internet already is that database, continuously updated and free to query. So version 2 extracts the key clues and runs a targeted web search, letting a reasoning model resolve the title from live results, the same goal at a fraction of the cost. The cheapest, most reliable system often reuses infrastructure that already exists.

/ Skills demonstrated

Mobile DevelopmentAI PipelinesRealtime (WebSockets)System DesignCost Engineering

/ Built with

React NativeExpoTypeScriptZustandPythonFastAPIPydantic v2RedisOpenAI (transcription + reasoning + web search)AWS RekognitionOpenCV (YuNet / SFace)faster-whisperyt-dlpffmpegSupabaseTMDbDockerRailwayGitHub Actions
Source on GitHub