SUNSETSEJONG UNIVERSITY
← Back to projects
P-03 · Flagship

MoodFlix

FlagshipPrototypeAI ExperimentWeb
MoodFlix multimodal movie recommendation research prototype. Face, audio, and context inputs feed DeepFace emotion analysis, Librosa audio features, and contextual signals into Ollama TinyLlama and a Random Forest model trained from logged data, enriched through TMDB, to produce recommendations. Dashed borders mark research-grade or incomplete modules.SIG-MFX · MULTIMODAL STACKINPUTSFacewebcamAudiomic / envContextsessionPROCESSINGDeepFaceemotionLibrosaaudio featuresContext parseHTTP serverCSV loggingsync pathsMODELSOllamaTinyLlamaRandom Forestlogged dataMerge & rankprototypeOUTPUTTMDBenrichmentRecommendReact clientclient + TMDBresearch modulesfragile paths
System diagram

Verified summary

Multimodal movie-recommendation research prototype combining visual, audio, environmental, and model-generated context.

Problem or experiment question

Can facial emotion, audio features, and local model context improve movie recommendations in a research setting?

What was actually built

  • React client with webcam, environment, and audio capture direction
  • Python raw HTTP server endpoints
  • DeepFace facial-emotion analysis and Librosa audio-feature extraction
  • Local Ollama/TinyLlama usage with Random Forest component from logged data
  • TMDB enrichment and CSV-backed logging paths

Important technical decisions

  • Raw HTTP server keeps the research stack lightweight
  • CSV logging supports offline model experimentation
  • Local Ollama path avoids external model hosting during development

Architecture or workflow

  • Client capture modules send multimodal payloads
  • Server orchestrates emotion, audio, and model inference
  • Recommendation layer merges logged features with TMDB metadata

Known limitations

  • Missing or fragile imports were observed in the codebase
  • CSV merge and schema handling is brittle
  • Wildcard CORS and manual request parsing reduce reliability
  • Research-prototype quality, not production-service quality

Repositories

Related projects