Hello World: Building This Website
A first post about building my personal website with Next.js, MDX, and modern web technologies.

I like understanding how things work and then trying to make them work better. Mostly around search, retrieval, and agent-based systems.
I spend most of my time thinking about how we find and use information.
That usually means working on search and retrieval systems — from lexical and semantic search to rerankers, and evaluation. More recently, I've been exploring agent-based approaches, where systems don't just retrieve documents, but reason about what to retrieve.
I'm especially interested in the parts that are easy to gloss over: assumptions, trade-offs, evaluation metrics, and why something works (or doesn't).
Hybrid search, filtering, and ranking pipelines that behave well in production.
Embeddings, rerankers, and evaluation loops that keep results honest.
Agentic retrieval workflows: tools, skills, and multi-step intent.
A large part of the software we rely on today exists because people decided to build things in the open.
I actively contribute to open-source projects and try to build my own tools the same way. If you're curious about search, retrieval, or agents, contributing to open source is one of the best ways to learn — and to give back to the community that makes most of this work possible.
Around-the-clock contributor to one of my favourite open-source projects: Sentence Transformers.
High-throughput LLM API orchestration for running many requests reliably and efficiently.
Focus: Performance, reliability, and developer ergonomics.
Structured outputs from language models — predictable, schema-driven results instead of fragile text parsing.
Knowledge distillation for BERT-style models, inspired by MiniLM-style approaches — smaller, faster models with strong performance.
I write to understand things better and to share what I learn along the way.
If something here was useful, interesting, or confusing — feel free to reach out.