LLibrarymindsinlibraryminds.hashnode.dev·3d ago · 24 min readSemantic Search Over Transcripts in Python: What Embeddings Fix and What They Don'tShort answer: keyword search finds the words you typed, and vector search finds paragraphs that sit close to them in meaning. I built both in about 120 lines of Python, ran them on a made-up talk tran15KA
IKIbrahim Kedir Bedruinbinasmart.hashnode.dev·3d ago · 4 min readWhy our Amharic AI searches before it speaksHow we built Bini, an assistant that answers Ethiopians' everyday questions in Amharic, and what we learned about the real cost of the Ge'ez script. Ask a general-purpose AI assistant how much it cost01I
MSMinaruzzaman Shovoninminarsvn.hashnode.dev·6d ago · 7 min readI Trained an Open Decision Model on a Free GPU in 46 Minutes. On My Tests, It Beat Laya.TYPIC answers typed questions in a single forward pass, with no text to parse. Here's how I built it, what worked, and where it still falls short. A lot of software now asks an AI model tiny questions00
RRhymeUinrhymeu.hashnode.dev·Sep 27 · 8 min readDeclining names in poem titles: lessons from generating poems in Slavic languagesBy the RhymeU team, independent developers. We build RhymeU, a free generator of personalized poems in 29 languages; this post is about a bug we shipped and the problem space behind it. Here's a title00
RSRuvarashe Shalom Nemarambainruva.hashnode.dev·Sep 26 · 10 min readCan I Learn NLP in a Week?Welcome to another blog post where we come together to create a 7-day roadmap for our iconic 7 day Natural Language Processing Sprint. So, what is NLP? Formally, Natural Language Processing (NLP) is a00
MAMuhammad Ariel Shakaramiroinshaka-ai.hashnode.dev·Sep 25 · 8 min readOne Answer, Four Different Scores: Learning LLM Evaluation the Hard WayThis week's material was about evaluating LLMs: offline metrics (tested against a fixed dataset) versus online metrics (measured from real user interaction). It sounds straightforward on the slides. O00
NSNagappan Sinnagspidey.hashnode.dev·Sep 24 · 8 min readTokens & Tokenization: Why AI "Counts" Words WeirdlyYou paste a paragraph into ChatGPT or Claude, hit enter, and somewhere a meter is ticking. Not a word count. Not a character count. Something else — "tokens" — and the number never quite matches what 01B
MYMohammed Yahya Ahmedinlearningwithyahya.hashnode.dev·Sep 22 · 5 min readChoosing the Embedding Architecture for BrainZenWhen I first started adding semantic search to BrainZen, I figured the actual vector search algorithms would be the biggest headache. I was wrong. The very first hurdle was a much more foundational qu10
SKSpeed Kinyasbd-lib.hashnode.dev·Sep 22 · 16 min readyasbd-lib v1.0.0: how beta finally endedBefore 0.12.0, yasbd-lib was a promise. 39 languages, spaCy and pysbd integrations, a CLI, a streaming API, a benchmark suite. Every one of those language profiles took real time to build because we w00
MYMohammed Yahya Ahmedinlearningwithyahya.hashnode.dev·Sep 21 · 5 min readBuilding Search for BrainZen: From Embeddings to Hybrid RetrievalSearch initially seems simple: type a query, get matching content from the database. But once I started trying to make BrainZen—my second-brain app—actually understand what I'm looking for by meaning 24M
