feat (umap, clustering): Added UMAP projection and clustering benchmark pipeline

[Harsh16gupta]

This PR introduces a clustering benchmark pipeline to evaluate different grouping strategies for note categorization. High-dimensional 384D note embeddings are optionally reduced using UMAP before running the clustering algorithms side-by-side.

Fixes: #15

Implements:

• UMAP projection (384D to 10D) using umap-js.
• Custom K-Means and K-Medoids implementations.
• Wrapped HDBSCAN with support for cosine distance and noise tuning.
• Silhouette coefficient scoring to evaluate cluster quality.
• Shared Mulberry32 utility for deterministic, reproducible results.
• Benchmark integration to log a comparison table of strategies in the console.

test

[HahaBill]

@Harsh16gupta Hi Harsh, I created this GitHub issue so we can have more discussion regarding the clustering algorithms, considering other ML approaches, and possible improvements. This is the new GitHub issue that I created: #15

I think it's better to report your experiments and findings, our discussions, etc. there than here.

[Harsh16gupta]

Hey Bill, thank you for creating an issue to tackle this.
I have deleted that long testing results and updated it in the isssue. Will continue sharing my findings there.

[Copilot]

Pull request overview

Adds an end-to-end clustering benchmark pipeline for note-embedding vectors, optionally applying UMAP dimensionality reduction before running multiple clustering strategies and scoring them via silhouette coefficient. This supports Issue #15 by enabling side-by-side comparison of grouping approaches for note categorization.

Changes:

• Introduces deterministic benchmarking utilities: Mulberry32 PRNG, UMAP projector wrapper, and distance/quality metrics (incl. silhouette score).
• Adds clustering implementations/wrappers (K-Means, K-Medoids, HDBSCAN) plus a benchmark runner that logs a comparison table.
• Integrates the benchmark run into the existing “Test Embedding” command and adds new dependencies (umap-js, hdbscan-ts).

Reviewed changes

Copilot reviewed 11 out of 13 changed files in this pull request and generated 5 comments.

Show a summary per file

File	Description
tsconfig.json	Enables skipLibCheck for faster/less strict library type checking.
package.json	Adds umap-js and hdbscan-ts runtime dependencies.
package-lock.json	Locks newly added dependencies and their transitive tree.
src/utils/prng.ts	Adds Mulberry32 deterministic PRNG utility.
src/types/projector.ts	Defines UMAP projector option types.
src/types/cluster.ts	Defines clustering config/strategy/result types for benchmarking.
src/pipeline/UmapProjector.ts	Wraps umap-js with deterministic seeding and metric selection.
src/pipeline/clustering/metrics.ts	Adds cosine/euclidean distance and silhouette scoring.
src/pipeline/clustering/kmeans.ts	Implements K-Means with k-means++ initialization.
src/pipeline/clustering/kmedoids.ts	Implements K-Medoids (simplified PAM).
src/pipeline/clustering/hdbscan.ts	Wraps hdbscan-ts with cosine support via normalization.
src/pipeline/clustering/benchmark.ts	Runs strategies, computes scores, and logs a comparison table.
src/commands/testEmbed.ts	Runs the clustering benchmark after embedding and logs cluster assignments.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[HahaBill]

All looks good for now. Thank you for this PR :)