Small LLMs commit to a hunch and stop tracking candidates. This post breaks down the failure modes, describes a loop that grows a creator and a solver in turn, and reports how far a per-turn reward actually moved the numbers.
[Read More]
When the Timeline Fills Up with Reposts and Copies: Designing a Filter That Only Changes What You See
How to fix a timeline buried under reposts and re-uploaded images without deleting anything: why mute and keyword filters fall short, how to pick the original by timestamp, and how to keep a way back when the filter guesses wrong.
[Read More]
Folding the Search for a Repost's Origin into One Query
Tracing a reposted image back to its original post is a manual loop across several reverse image search engines and dozens of page dates. How Origin Finder collapses that loop, and what it still leaves to a human.
[Read More]
Turning One Image into a Talking 3D Character: Making the Build Repeatable
Full mesh generation gives a different character on every run, and hand editing leaves no record. How driving VRoid Studio through an MCP turned character work into a repeatable cycle of change, export, compare, and accept.
[Read More]
Reading Trend Performance in Numbers — Filling the Blind Spot in View-Count Rankings
A view-count ranking cannot tell you whether a trend video worked. This is how one app puts a viral coefficient, a median, and an engagement rate on one screen, aligns the comparison group, and where those numbers still fail.
[Read More]
Stop Calculating Business Trip Travel Time by Hand
Copying addresses into a map app and subtracting travel time from every meeting is a nightly ritual for anyone who travels for work. BookingButler derives the origin from the calendar and works backwards from arrival. Here is what got automated and what stayed manual.
[Read More]
A Search-Enabled Chat UI in the Shortest Path: A Research Assistant with Streamlit and a ReAct Agent
Ask a chat model about last week and you get stale knowledge. Then you search in another tab, paste the result, and ask again. Here is how a LangChain ReAct agent with Google and Wikipedia tools collapses that loop into one Streamlit file, and where the approach stops working.
[Read More]
Cutting the Cost of Knowing What Is Trending on X
Following an X timeline costs three things: finding the right posts, reading them, and summarizing them. How TopicDigestX moves each to a server, the trade-offs, and what is still not built.
[Read More]
Fine-tuning a Japanese LLM Without a GPU: One Mac, One Pipeline
No CUDA GPU, no cloud quota, no separate evaluation scripts. How FT_LLM turns fine-tuning a Japanese LLM into one Mac workflow, and where the approach still falls short.
[Read More]
Three Constraints in Setlist Planning, and How Each One Was Removed
Planning a setlist where members switch instruments runs into three constraints: transition cost, stage time, and where the break goes. How the manual limits were framed, and how OR-Tools TSP plus a break inserter removed them.
[Read More]