May 2023 – Dec 2023 · 7 mos · Remote

• Search projects and priorities, including new LLM integration efforts, website indexing (hybrid retrieval with keywords + vector) and related search applications and maintenance. Developed and maintained the company's enterprise search, and worked on migrating the search-stack to use Semantic Search from Lexical Search using SOTA techniques.
• Designed an enterprise RAG pipeline with embeddings following Vector DB sharding and replication strategies, e.g., chunk documents optimally, avoid semantic drift, etc.
• Built a retrieval layer with hybrid search (BM25 + dense) + rerankers for better precision.
• Designed a caching strategy for 'hot queries' for latency trade-offs vs. retrieval accuracy.

May 2021 – Apr 2023 · 1 yr 11 mos · Remote

• Working in the Document Technology team on website indexing (keywords), boosting relevance for search queries (TF/IDF), documentation search-engine optimization, and other related search applications, bugs and maintenance.
• (shipping high-performance, scalable code in Java, C, Python and/or Wolfram Language; build tools: TeamCity, Jenkins, Gradle; version controls: git, JIRA)

Aug 2020 – Apr 2021 · 8 mos · Remote

• Working on several R&D projects related to cellular automata, symbolic computation and computer algebraic systems under the Special Projects department.

Feb 2021 – Oct 2021 · 8 mos · Bengaluru, Karnataka, India · Remote

• IBM Global Remote Mentee Program
• Project: Worker-node model for MP-SPDZ (Applied Cryptography)
• Mentor: Dhinakaran Vinayagamurthy
• Objective: To develop a framework for N parties who want to jointly compute a function with their individual inputs while keeping their inputs secret, and also want the final outcome to be differentially private, i.e, a noisy result instead of providing the actual result.
• Expected outcomes: Prototype, paper
• Actual outcomes: One of the 11 global contributors of IBM's Differential Privacy library; changed upper bound of variance from (upper - lower) ** 2 to ((upper - lower) ** 2) / 4 guaranteed by Popoviciu's inequality (https://github.com/IBM/differential-privacy-library/pull/45).
• Research readings included MPC protocols (SPDZ, SCALE-MAMBA, MP-SPDZ), DP mechanisms (Laplace, Gaussian, Distributed DP), and studying benchmarks of open-source privacy libraries (IBM diffprivlib, Google Privacy on Beam, OpenMined, Opacus, OpenDP).

May 2019 – Jul 2019 · 2 mos · Kolkata, West Bengal, India · On-site

• Summer Research Intern at R.C. Bose Centre for Cryptology and Security powered by Microsoft Research India.
• Details:
• Initially did a reading project under Dr. Mridul Nandi on AES and it's modes of operations(special attention to Galois Counter Mode of operation).
• Then, worked under Dr. Nilanjan Datta and Mr. Ashwin Jha on Differential and Linear Cryptanalysis of AES and related block ciphers.
• Wrote a general code to form an LP file which will serve as the instruction for GNU's GLPK tool to find the number of Active S-Boxes for any given rounds of ANY GIVEN Block Cipher,which can be considered as an updated code that N Mouha described in his paper Differential and Linear Cryptanalysis using Mixed Integer Linear Programming for only AES(https://www.esat.kuleuven.be/cosic/publications/article-2080.pdf)
• Wrote a generalized optimized code to find the branch number of any given block cipher, and then by using GLPK solver find out the number of active S-boxes for any rounds of any block cipher and hence using mathematical relations determine whether the cipher is resistant against Differential and Linear Cryptanalysis or not.
• Result: This code was hence built with the intention of making it work as an automated solver for any block cipher to determine whether it is resistant to linear and differential cryptanalysis or not. With certain precomputations, the optimised code worked somewhat within 2 seconds and gave correct result.
• Medium Blog Article: https://medium.com/@justanotherlad/generalized-linear-and-differential-cryptanalysis-check-for-any-block-cipher-r-c-bose-bac010b7d6e6