Data · Independent Research · Learning in public

A small place for the things I’m learning, building, and still figuring out.

This is a new space for me. Part portfolio, part notebook, part exploratory blog. I want it to feel light, honest, and alive while I write about data engineering, independent research, and the odd bit of vibe coding.

Data systems Independent research Vibe coding Learning notes
Read notes Independent research

About

An introduction, without turning the page into a résumé wall.

Personal Notebook · Learning In Public

A little about me

My name is Tanvi. I work as a Lead Data Engineer, mostly around financial services data systems, SQL, SSIS, ETL, governance, migration work, and the practical details that make data platforms usable over time. More recently, I have also started working with dbt and Databricks. Alongside that work, I pursue independent research into language-model reliability and decoding.

This site is a new thing for me. I want it to grow slowly as a place for notes, early independent research, small experiments, and the kind of learning that does not always fit neatly into a formal project description.

Independent Research

My first independent research thread is still early, but I want to document it carefully.

This section is for independent research ideas, preprints, questions, and the parts that are still uncertain.

First preprint

Adaptive Context-Aware Embedding Filtering

Large Language Models (LLMs) can generate fluent and coherent text, yet remain prone to producing factually unsupported outputs. We present Adaptive Context-Aware Embedding Filtering (A-CAEF), a lightweight decoding-time method that adjusts token selection using context-aware drift signals. The method does not require retraining, external retrieval, or changes to model parameters.

Notes / Ideas / Blogs

Notes, ideas, rough drafts, and small learning trails.

A lightweight notebook for things I am still shaping, not a finished library.

Draft

The Geometry of Generation

May 2026 · Independent Research

A draft reflection that shifts the focus from geometry of representation to geometry of generation.

  • LLMs
  • Decoding
  • Reliability
Open note

This place is meant to stay a little unfinished.

That is the point. I want the site to grow with the work: some parts researched, some parts drafted, some parts simply explored.