All posts
PythonFlaskNLPscikit-learnspaCy

ChefMate: NLP + ML Pipeline from User Input to Recipe

November 8, 20256 min read
Share this post on XShare this post on LinkedIn
Noisy ingredient text passes through normalization, fuzzy matching and recipe ranking to produce suggestions.
A simplified conceptual illustration of ingredient normalization and recipe recommendation. The examples illustrate processing stages rather than measured model outputs.

ChefMate: NLP + ML Pipeline from User Input to Recipe

ChefMate is an AI-powered recipe suggestion app I built as a solo developer.

Problem

Ingredient input is usually messy: spelling errors, synonyms, and inconsistent naming.

Pipeline

  1. Normalize ingredients with spaCy
  2. Handle input variation with FuzzyWuzzy
  3. Predict cuisine and recipe candidates with scikit-learn
  4. Return filtered suggestions through a Flask app

Dataset

I trained models using a 643MB recipe dataset and focused heavily on preprocessing quality.

Result

The NLP + fuzzy matching pipeline improved ingredient recognition by around 70% on noisy input scenarios.

Takeaway

For practical ML products, input normalization often drives quality more than model complexity.

More Posts

Neuro-Symbolic AI: Why the Next Real Step Isn't a Bigger Model

Deep learning sees patterns; symbolic systems follow rules. Neither alone is enough for reliable reasoning. A practical look at why hybrid neuro-symbolic systems, from AlphaGeometry to tool-using LLMs, are quietly solving problems that pure scaling cannot.

The Reversal Curse: Your LLM Knows "A is B" but Not "B is A"

Train a model that Tom Cruise's mother is Mary Lee Pfeiffer, and it still struggles to say who her son is. A 2023 finding shows large language models store facts in one direction only, and it says something deep about what knowing even means for a transformer.