Currently Cambridge, MA
Doing M.S. + A.B. Computer Science · Class of 2026

Vlad
Cainamisir

Data Systems · AI

Research and software. I build things and make things go brr
Mostly in/around the Data and AI Systems Lab.
Formerly at DRW, TileDB, Bloomberg.
Concurrent B.A./M.S., Harvard University.

§ 1 About Me

I'm originally from Bucharest, Romania. I moved to the US in 2022 to study CS at Harvard.
I love thinking about data systems and AI.

I recently started sailing at MIT, so you might see me there once the Charles unfreezes lol.

I love reading and here are some of my current favorites:

  • The Master and Margarita Mikhail Bulgakov
  • The Making of Prince of Persia Jordan Mechner
  • The Maniac Benjamín Labatut
  • The Making of Karateka Jordan Mechner
  • Surely You're Joking, Mr. Feynman! Richard Feynman
  • Kitchen Confidential Anthony Bourdain
Location
Cambridge, MA
Status
M.S. + A.B. · 2026
Contact
email github linkedin
§ 2 Experience
DRW
Research Engineer Intern · Chicago, IL
Jun — Aug 2025
view →
  • Compiled Python to native C++ speed, removing the interpreter from latency-critical paths entirely. Added numpy, arrow, and pandas support.
  • Bidirectional low-latency Python↔C++ interoperability layer eliminating data copying and (de)serialisation overhead for high-frequency cross-language execution.
  • DRW's fastest volatility forecasting service: backtesting and predictions in <100ms for a full year of data. Improved R² by 0.005.
TileDB
Software Engineer Intern · Cambridge, MA
May — Aug 2024
view →
  • Resource scaling strategy for the distributed execution graph during vector ingestion: 8× memory reduction with only 4% latency increase.
  • Type erasure integration of inner product and cosine similarity without performance compromise.
Bloomberg L.P.
Software Engineer Intern · New York City, NY
Jun — Aug 2023
  • Highly available interest rate data system handling 2–3 trillion queries per day. XSD schema became the company standard for interest rate indexes.
  • Local caching system: throughput thousands → 80M queries/sec; latency 10ms → 20μs.
§ 3 Education
Harvard University
2025 — 2026
Master of Science in Computer Science
Harvard University
2022 — 2026
Bachelor of Arts in Computer Science
Teaching Fellow — CS 61 Systems Programming · CS 165 Data Systems · CS 205 High Performance Computing
Coursework — Operating Systems · Networking · HPC · Big Data Systems · Distributed Systems (MIT)
§ 4 Recent Research
P.01

FreeThrow

C++, Python

Learned predictions accelerating HNSW vector search indexes by 2×–3× at zero overhead — via embedding model tuning and graph decomposition.

Vector Search · Learned Indexes
view →
P.02

LayoutDB

C++

ANN index with hybrid on-disk/in-memory design: small vector prefix in DRAM, bulk on NVMe. 100× speedup over IVF_FLAT via adaptive projections (Johnson–Lindenstrauss), minimal data movement, smart query plans, and SIMD kernels.

Vector DB · Storage Systems · SIMD
view →
P.03

Retrieval-Augmented Models

Python · PyTorch · vLLM

Dynamic LoRA expert retrieval for continual learning. Embedding-based routing selects the nearest specialist adapter at inference time: +10% absolute / +25% relative over Qwen 2.5-Coder on CodeFlowBench.

LLM · PEFT · Continual Learning
view →
P.04

Agentic Browser Exploits

Perplexity Comet, BrowserBase

Exposed multiple zero day exploits in Perplexity Comet and BrowserBase. I have reported them with reproducible instructions. Writing a paper on suggesting architectural changes to Agentic AI attention mechanisms to mitigate prompt injection attacks.

Details coming soon after patches are released
§ 5 Technical Profile
Languages C++ · Python · JavaScript / TypeScript · P4 · HTML/CSS · Verilog
Domains Database internals · High-performance computing · ML Systems
Competitive Programming Codeforces Candidate Master · National Informatics Olympiad, Romania
§ 6 Contact
vcainamisir@college.harvard.edu · +1 857 706 9103 · github.com/cainamisir · linkedin.com/in/cainamisir