Interactive demos.

Small working illustrations, one per role and one per project write-up. They run in your browser, send nothing and store nothing. Where the real data is private, they use made-up sample data and say so.

Roles

• DHS

  Pick a fiscal year from FY2019 to FY2025 and compare the office's enacted budget with FY2018, as enacted and counted on the same functions. Every figure comes from the documents listed under Sources.

• FAA

  How many people leave, and how many must be hired to hold headcount, as separation and retirement rates change. A made-up workforce; not FAA data and not the original model.

• Department of Veterans Affairs

  Score a made-up team against the skills a position needs, the matrix idea behind a skill-matching model. Not VA data.

• Federal Protective Service

  Turn countermeasures on and off for a sample facility and see how well each threat is covered. A made-up matrix; not the Interagency Security Committee method and not FPS data.

• NASA

  How hiring and attrition move headcount over ten years, and how long a hiring class stays. A made-up organization; not NASA data.

• U.S. Census Bureau

  Type a few lines of SAS and see the matching Python (pandas) code. A teaching converter for a small part of SAS.

Projects

• Analyzing Problems and Solutions 3/4

  Rank fixes for the five Lighthouse metrics with a weighted decision matrix: move the weights and the order changes. Made-up scores.

• The AI Business chatbot project

  Ask a question about a made-up shop: TF-IDF finds the closest known question, and the page shows the SQL, the result and a plain answer.

• Churn Predictor Project

  Change a made-up customer's details and contract type and see the predicted chance of leaving, from a logistic model with made-up coefficients.

• Cluster Market Project Report

  Group 160 made-up customers with k-means, re-run from new starting points, and read the elbow chart for choosing the number of groups.

• EmotionVision: AI-Driven Emotion Recognition in Facial Images

  A confusion matrix for seven emotions, and how raising the confidence threshold trades coverage for accuracy. Made-up predictions.

• FaceAI Project Report

  Similarity scores for genuine and impostor pairs, a decision threshold, and the false accept and false reject rates it gives. Made-up scores.

• Feasibility and Required Tools 4/4 – Conclusion

  A learning curve: a hand-written rule against a nearest-neighbour model as the number of examples, the label noise and the variety change. Made-up cases.

• Human Classification (Part 1)

  Group made-up users with k-means and step through the rounds as the groups settle.

• Human Classification (Part 2) These three phases are as follows

  The three phases in order: tag a page, build a user's profile from visits and follows, then match the user to a class. Made-up pages and users.

• HumanClassification (Part 3) – Parsing the page

  Parse sample pages: structured data read first, then the title, heading and meta tags, with weighted keyword counts. Made-up pages.

• NewsClassifier: Automating News Categorization

  A naive Bayes classifier trained in your browser on 42 made-up headlines: type a headline and see the category probabilities and the words that decided it.

• ObjectDetector Project: A Deep Learning Approach

  Intersection over union: move and resize a predicted box against the true box and see whether it counts as a hit at 0.5 or 0.75.

• PDF Chatbot

  Split a made-up guide into chunks, change the chunk size and overlap, and see which chunk best matches a question.

• Sas2py – Migration is the only way from SAS to Python

  Type a few lines of SAS and see the matching Python (pandas) code. A teaching converter for a small part of SAS.

• SmartChatbot Project

  Match a message to made-up support intents by similarity, and set how sure the bot must be before it answers instead of asking again.

• NLP Tools Comparison: SpaCy vs. NLTK

  Two tokenizing rules side by side on your own text, with the places they differ marked. The rules illustrate the kind of difference the write-up compares; they are not spaCy or NLTK.

• Managing Big Data with Spark and Hadoop: Performance and Applications

  Count words the MapReduce way: step through Map, Shuffle and Reduce, and choose how many partitions the work is split into.

• Stock Predict project report

  Short and long moving averages over a made-up price series, with crossovers marked. An illustration only, not investment advice.

• SupplyForecast project report:

  Forecast made-up monthly demand with simple, Holt or Holt-Winters smoothing and see the error on twelve held-out months.

• What is MLOps and Why is it Important?

  Watch for data drift: move a model input's live data away from its training data and see the population stability index go from stable to act.

More

• Nutrition analyzer

  Build a plate from a few foods and see its energy and nutrients against the Daily Values on US nutrition labels. Values per 100 g from USDA FoodData Central; general information, not dietary advice.