Open to working student and internship roles
Hey, I am Aman
data engineering, databases, and ML for biology. always building from scratch.
I'm a research intern at Prof. Schliep's medical bioinformatics lab and an MSc Artificial Intelligence student at BTU Cottbus-Senftenberg in Germany. At the lab I'm working on fast embedding-based retrieval of binding sites for oligonucleotide drugs.
Before that I spent almost four years as a data engineer at Infosys on Apple's data platform, owning 300+ production pipelines from SAP HANA into Oracle and Snowflake. In my spare time I'm writing a storage engine in C++ to see how databases work underneath.
Explore my work
Lab workNowResearchOligo binding-site retrieval at Schliep LabView research →Industry3 yr 10 moExperienceData engineering at Infosys for AppleView experience →Open source10ProjectsStorage engines, warehouses, PCAView projects →Reading list20PapershelfIndexes, retrieval and DNA modelsView papers →Essays4WritingDatabase internals and linear algebraView writing →Practice600+LeetCodeAlgorithm problems, 200+ in C++View profile →
Recent projects
All projects ➔Things I built to understand how they work.
- ★ 1miniDBin progressA B+Tree-indexed key-value storage engine built one layer at a time. Each layer fixes a problem the one below it causes: append-only log, then an in-memory hash index, fixed 4 KB pages, an LRU buffer pool with pin counts and dirty tracking, and finally the B+Tree.
- ★ 1drinkmix-data-warehouseA star-schema data warehouse and ETL pipeline for a beverage retail chain (BTU course project).
- PCA-Image-CompressionImage compression with PCA implemented from first principles, no PCA libraries.
- Oligo binding-site retrievalresearchResearch prototype from my internship: a two-tower encoder trained on primer3 binding energies.
On my papershelf
Full papershelf ➔What I'm reading right now.
- Hist-Tree: Those Who Ignore It Are Doomed to LearnCrotty · CIDR 2021 · Presented it in my MSc seminar.
- The Case for Learned Index StructuresKraska et al. · SIGMOD 2018 · The paper Hist-Tree argues against.
- The PGM-index: a fully-dynamic compressed learned index with provable worst-case boundsFerragina & Vinciguerra · PVLDB 2020 · One of the learned indexes Hist-Tree beats.
- Accurate prediction of oligonucleotide probe hybridization efficiency via a transformer-based modelPMC
- DNABERT-2: Efficient Foundation Model and Benchmark for Multi-Species GenomeZhou et al. · 2023
Recent writing
All writing ➔Short essays on LinkedIn.