Experience
-
- Researched on how much specification Large Languages Models require to generate correct
& safe code with minimum hallucinations along with that for testing the rust code generated by LLMs we
built a software, under the guidance of Professor Pratik Jawanpuria from IIT Bombay and Dushyant Rajput, Nirdesh Chauhan from AltSlate.
- Evaluated AI assisted migration of 30 GNU Core Utilities from C to memory safe
Rust across 5 specification levels, 5 retry limits
(total 750 experiments) using an automated differential testing & statistical
analysis bootstraping.
- Identified the specification vs cost tradeoff and determined the optimal specification level
(knee of curve) beyond which additional guidance yields negligible improvements in
quality of code generated by LLMs.
-
- Worked on improving the performance and scalability of the ATDM
(Aggregate Travel Demand Modeling) tool for Bengaluru's transportation network under the guidance of
Professors Abdul Rawoof
Pinjari and Tarun
Rambha.
- Implemented memory and runtime optimizations, leveraging multithreading and parallel computing to reduce
execution time from several hours to minutes. Optimized numerical
stability and memory usage to support 3,000+ transport zones, while resolving performance
bottlenecks and crash issues in the previous Python-based implementation, significantly enhancing
scalability and reliability.