Hey there 👋

I'm Gaurang, working as Research Engineer at Raytheon Technologies. I work on language model and autonomous agents. My broad research interests include language model, agentic systems and AI4Code

PS: I enjoy implementing things from scratch 🛠️

Blogs

Continuous Latent Space

:linkedin_icon-svg: LinkedIn | :1707226109new-twitter-logo-png: Twitter | :githublogo: GitHub

Why this website?

I created this website to document my research, learning journey, and anything else I find intriguing. The notes here span a range of topics: language models, agents, and deep learning to space technology and satellite imagery. While the blogs are still a work in progress, you can read through [public_cache] which is sort of like my L2 cache (but updated more frequently)

• Design Pattern v2
• Positional Encoding through Qs
• BackPropogation: Math & Intution #DL
• Neural Collapse & Long-Tail in NN #DL
• Agents: Architecture and Imlementation (from scratch) #llm
• Do DL models overfit? #DL
• Controls: AODS & Thrust Vectoring for sounding rockets #add #misc
• SAR: Interpretability & Fusion #add #misc

Previous Experience

1. Research Intern @ Raytheon Technology:
   • Developed a knowledge summarization system to curate, collect and transform scientific information from heterogenous, unstructured data using agentic pipelines capable of text classification, mathematical calculations and structure knowledge representation while incorporating human-in-loop responses leading to projected annual cost saving of 800,000 USD.
   • Rearchitected internal GenAI platform to Event Driven workflow to support Model Context Protocol (MCP) showcased usecase to accelerate project management tasks on the JIRA and GitHub platforms.
   • Designed a modular, LLM-driven C++ to Rust translation pipeline integrating AST parsing, repository indexing, vector and graph based code representation, compiler-level validation, and agent-based refinement. Achieved a 60–80% reduction in manual rewrite effort, enabling scalable legacy code modernization and generation of idiomatic, production grade Rust across entire codebases.
2. Project Intern @ ISRO (URSC):
   • Developed an LSTM-based dynamic routing algorithm to enable Inter-Satellite Link (ISL) capabilities for ISRO’s LEO and MEO satellite constellations, significantly enhancing network scalability and reducing communication latency.
   • Proposed a 6 satellite constellation architecture for a Lunar Earth communication system, achieving over 80% line of sight coverage and full coverage of lunar south pole.
   • Mission Support for the Aditya-L1 mission.
3. ML Intern @ LIGO India (IUCAA Pune):
   • Conducted research at the Inter-University Centre for Astronomy and Astrophysics (IUCAA) under the LIGO-India initiative, applying deep learning to detect gravitational waves from binary black hole mergers.
   • Benchmarked the performance of DL models for GW detection, comparing temporal (RNN, LSTM) and spectral (CNN, ViT) approaches across time and frequency domain representations.

Other Achievements

1. Hold 2 trade secrets for my work at ISRO.
2. Co-founded Space Research Group STeRG at university (MIT Pune).

Contact

I'm excited to discuss/collaborate on any ML, Math, Computing, or Space related topics. The best way to reach out to me is via [email protected] but you can also find me on other platforms :linkedin_icon-svg: LinkedIn | :1707226109new-twitter-logo-png: Twitter | :githublogo: GitHub