MyBlog

I have made a pretty meaningful update to this website: the blog engagement system is now powered by urthreads, my self-hosted static-site engagement service for Cloudflare Workers and D1.

I came up with the idea of this project during the development of this website. I wanted to add engagement and reactions to my posts without having to maintain a user flow that would be mostly unnecessary since I’m the only author of the blog.

While I was looking for ready services to implement my idea, they were either not working properly on the free tier, or it needed a subscription tier to use. So I made it open source and free, hosted by you.

What Changed

• Likes and comments now run through the urthreads Worker.
• Comments are still moderated before they appear publicly.
• The Worker keeps using Cloudflare D1, so the site itself remains static and GitHub Pages-friendly.
• The backend can now support richer features like threaded comments, comment likes, denied keywords, stats, and audit logs.
• The blog feed now has a wider responsive desktop layout while keeping the mobile layout compact.
• Posts with media now use a two-column desktop feed layout, with the text continuing full-width after it passes the media preview.
• Single images and image sliders are extracted into consistent media previews on the blog feed.
• Image sliders can be dragged or swiped on mobile and tablet viewports.
• Comment spacing on the feed is tighter, and the extra divider between post content and comments has been removed.
• Homepage featured-post excerpts now keep post headings as compact bold labels instead of rendering them as full headings.
• Code blocks are more compact and now use a blue theme that better matches the rest of the site.

I also added a few smaller improvements around the site while making the change. The browser tab now has a small theme-aware My SVG favicon, the blog feed avatar has a clearer blue ring, author labels now read more naturally, and post likes stay synchronized across repeated buttons for the same post.

Posts can now include image sliders using ordinary Markdown images, which makes it easier to write visual project updates without making every post layout custom.

Why urthreads

urthreads is meant to sit in the middle between a completely static website and a full backend application. It gives static sites just enough interaction to feel alive, but still keeps the deployment model simple: a Worker, a D1 database, and static dashboard files.

For this portfolio, that means I can keep writing posts in Markdown and publishing through GitHub Pages, while the interactive parts live in a focused service that can evolve independently.

What Comes Next

I want to keep improving the dashboard and moderation flow, especially around threaded conversations and post-level analytics. I also want this website to use the new media slider more often and keep refining how media-heavy posts are presented across the feed and individual post pages.

This update is mostly infrastructure, but it should make the blog easier to maintain and more pleasant to use over time.

Check out the urthreads Repository

I have recently released the first public version of urthreads; an open-source, self-hosted engagement service for static websites. It adds likes, moderated comments, threaded replies, comment likes, and a dashboard without needing to run a traditional backend server.

urthreads is self-hosted software for adding engagement features to static websites. Simply deploy the Worker to your own Cloudflare account, connect it to your own D1 database, and manage the dashboard with your own admin key. It adds page likes, threaded comments, moderation, dashboard analytics, and admin tooling for blogs, portfolios, documentation sites, and other static pages that need interaction without running a traditional server.

The goal is to keep the main website simple while moving interaction into a small service that you control. New comments are pending by default, the dashboard supports moderation and analytics, and the CLI can help with setup, admin keys, sessions, comments, likes, stats, and D1 health checks.

I am also using urthreads on this portfolio now. The likes and comments on this blog were previously backed by a small manual Worker setup, but I migrated the existing D1 database into urthreads without losing the old likes or comments. That was one of the main reasons I wanted to turn this into its own package: I wanted a reusable service that could support this site and future static websites without copying the same Worker code around.

This project is still young, and I am expecting to keep improving the dashboard, moderation workflow, and deployment experience. urthreads is open source, and everyone is welcome to use it, test it, and contribute.

Check out the Github Repository

Check out the npm Package

I figured I’d finally take advantage of my of my GitHub Student Developer Pack benefits; I would have done it at least a year sooner if I hadn’t been buried with either college work, or internships if not both.

That’s what I used to tell myself, but ever since graduating, I still have the very same problem; not with my assignments from work or college but from personal projects. I see now I like my head buried in projects that I’m interested, it was never about the deadline.

Growing to be so interested and involed in so many different domains in the umbrella term we use when we say “tech” has been an endless well that has always been able to satisfy my curiosity. So I decided to use this website for sharing my many interests in tech that doesn’t find its way on GitHub, or LinkedIn.

What to expect

• Updates on my existing personal and open-source projects.
• Experiments, and tech news.
• Personal updates and opportunities relating to my career.

Thanks for visiting — more posts are coming soon.

Summary

Tensor Serve now supports configurable search profiles that let you optimize the search pipeline for your specific deployment scenario—from lightweight local machines to enterprise production servers.

Quick Start

Switch to a Profile with the CLI

# Lightweight (local/embedded deployment)
tensor-serve config set-search-profile lightweight

# Balanced (default, general purpose)
tensor-serve config set-search-profile balanced

# Production (enterprise servers, large-scale)
tensor-serve config set-search-profile production
```Y

### View Available Profiles

```bash
tensor-serve config search-profiles

By default, these commands update local config.json. Add --server http://localhost:8000 to read or apply the profile through a running Tensor Serve server’s REST endpoints.

Performance Notes

Latency (P99) by Profile

Query embedding:                  10-20ms
-----------------------------------------
Lightweight:
  BM25 Okapi:                     2ms
  FAISS Flat:                     5ms
  Total:                          7ms + 20ms embedding = 27ms

Balanced:
  BM25 Okapi:                     2ms
  FAISS Flat:                     5ms
  Reranking (lightweight):        50ms
  Total:                          57ms + 20ms embedding = 77ms

Production:
  BM25+:                          3ms
  FAISS IVF:                      2ms
  Query Expansion (PRF):          5ms
  Reranking (balanced):           100ms
  Total:                          110ms + 20ms embedding = 130ms

Quality (Normalized MAP@5) by Profile

Based on TREC benchmarks:

Lightweight:    ~0.65
Balanced:       ~0.75
Production:     ~0.85+

Backward Compatibility

Existing deployments continue to work with default settings (Balanced profile):

• All previous configurations are preserved
• Search behavior is identical to pre-optimization release
• No re-ingestion required
• Optional to upgrade to new profiles

No breaking changes. All new features are opt-in.

I have recently released the beta version of Tensor, a service that turns your local ZIM files into vector databases that provides semantic understanding of the contents of your ZIM archive to OpenAI-compatible AI models.

Combining keyword search and semantic search, Tensor helps produce more accurate and adequate responses for the data you have included in a vector database. Tensor offers a full workflow that is built to enhance the local AI experience, however there is also optional features (offline-first by default) features such as web search for time-sensitive information, and configuring API keys to use (currently working on MCP Server).

This project is still under development, and in the process of optimization. tensor-serve is an open-source project, and everyone is is welcome to use and contribute to this project.

Check out the Github Repository

Check out the PyPi Package