csg
Résumé ↗
Menu
← All projects

Recommendation systems / Recommendation prototype

Scalable RecSys Engine

Find the next film from the one you love.

Movie discovery through content similarity and KMeans clusters, with optional approximate search and persistent caching.

Python · Flask · scikit-learn · Annoy

View repository ↗

01 / THE PROBLEM

What needed solving.

A movie catalogue becomes useful when a viewer can move from a familiar title to relevant discoveries. The project combines title lookup, genre browsing and alternative recommendation paths in a Flask application.

02 / FOLLOW THE WORKFLOW

See how it fits together.

A simplified example of the documented flow. Sample outputs are authored illustrations, not live model predictions.

Scalable RecSys EngineIllustrative workflow

Illustrative query

“Interstellar”

Look up a movie title or genre.

Feature representation

Science fiction · Drama · Adventure

Use genres and prepared movie features.

Configurable recommendation paths

Content similarity / KMeans / optional Annoy

Choose similarity or clustering.

Conceptual output, not a live ranking

Ranked film suggestions → cached result

Return related movie candidates.

03 / ENGINEERING

The decisions behind it.

Offer multiple recommendation paths

Content features and KMeans clustering provide different ways to find related titles. Annoy is optional, with feature flags and lower-memory fallbacks.

Cache repeated queries

In-memory and shelve-based persistent caches reduce repeated computation. JSON fallback and indexed title lookups support repeat searches.

Make resource constraints explicit

A fast-start option loads a subset of the data. Feature flags can disable expensive content-based or approximate-search paths on smaller machines.

04 / EVIDENCE & LIMITS

Claims with context.

KMeans + contentDocumented recommendation approaches
Persistent cacheResults can survive service restarts

The README describes large movie-rating data and optional Annoy indexing. Its “25 million movie entries” wording does not establish 25 million distinct titles or a reproduced load test, so that number is not used as a portfolio achievement.

Current limits & next validation

No verified Recall@k, NDCG or latency comparison was established in the reviewed documentation. The README references a differently named upstream repository; personal contribution boundaries need clarification before claiming sole authorship or a measured scalability result.

05 / READ THE SOURCE

Go deeper.

Based on public repository documentation reviewed on 7 November 2026. Results are attributed to the project’s recorded evaluations; they were not rerun for this portfolio.

NEXT PROJECTSentient Ambient AI

TOOLS BEHIND THE WORK

Explore the stack. Drag the ribbon or select a tool.