FEATURED ARTICLE / TOWARDS AI
A closer look at the method.
SEARCH / ENGINEERING / EXPERIMENT
I combined dense and sparse vectors to search medical research.
A reproducible comparison of dense semantic search, sparse BM25 and reciprocal-rank fusion over 12,868 PubMed titles and abstracts from the TREC 2018 precision-medicine pool. Embeddings caught meaning, BM25 preserved exact terminology, and reciprocal-rank fusion brought the signals together.
12,868PubMed records
5zero-hit queries with hybrid ranking
SHORTER NOTES / SHARED IN PUBLIC
LinkedIn posts.
Field notes from building software, using AI thoughtfully and bringing developers together.
ENGINEERING
When meaning isn't enough for search
What changed when medical retrieval had to preserve exact terms as well as semantic similarity.
Read the post ↗AI + LEARNING
AI should keep you in the learning process
A note from hackathon mentoring: review, debugging and ownership should stay with the learner.
Read the post ↗MODEL TEST
One racing-game prompt, two models
A deliberately small GPT-5.6 versus GPT-6 Astra experiment—not a benchmark.
Read the post ↗BUILDING IN PUBLIC
Post the experiment
The model comparison unexpectedly crossed 100K views; a reminder that finished and shared beats over-polished.
Read the post ↗CREATIVE AI
AI video still needs a point of view
Tools can produce frames; story, direction, pace and reference choices still belong to the maker.
Read the post ↗06NOTE / PRANSHU BANSAL
HACKATHONS
What hackathon demos don't tell you
Looking beyond a polished demo to whether a team understands what it built.
Read the post ↗07NOTE / PRANSHU BANSAL
PERSONAL NOTE
Remembering why I started building
A reflection on ambition, pressure, and finding enjoyment in software again.
Read the post ↗08NOTE / PRANSHU BANSAL
PERSONAL NOTE
Does the code still feel like mine?
From hand-written code notes to AI-assisted development, and the desire to understand the code.
Read the post ↗