Contributed over 25000+ lines of working Python code; learnt the meaning of ownership & 10x programming
Researched A2A protocol released by Google and made additions on it to support secure collaboration. Using it, made PSI and Dark Pools (privacy preserving trading) demos featuring multiple agents
As other side projects, built full-stack demos (frontend and api-integrated backend) end-to-end to showcase various products of Silence Laboratories, especially inside Silent Compute. Also wrote code for enrichment and completion of excel sheets containing potential leads of Silence using Lusha, Apollo and AnyMailFinder API
Engineered robust Python FFI bindings for Rust cryptographic libraries being used in the multi-agent systems
Made crewai and n8n workflows integrating the multi-agent communication with A2A protocol & Rust functions
Performed thorough code reviews of teammates, collaborated across engineering and product teams, wrote clean and efficient code, and refactored multiple, critical deep-tech code segments to improve long-term maintainability
Hackathon submissions:
Worked in a team of 3 to deliver a working product in the G20 TechSprint 2025 hosted in South Africa, winning $30,000 as first prize for the problem statement - consumer-consented and secure data portability
Architected an enterprise-grade MCP server implementing secure Private Set Intersection and submitted it in the NANDA MIT hackathon
Pioneered an intelligent AI orchestration system for automated document enhancement, enabling seamless content retrieval and integration from heterogeneous external sources, reducing manual effort by 90%
Developed and fine-tuned a custom BERT classifier for complex image decision tasks, achieving 80.4% accuracy and outperforming GPT-3.5 by 58% (51% baseline), demonstrating superior domain-specific performance
Engineered a comprehensive multi-modal dataset by implementing large-scale Wikipedia scraping infrastructure, creating 10K+ high-quality image-text pairs in structured XML format
Implemented robust evaluation pipeline using DocNLI framework for text entailment assessment and delivered production-ready GUI application in PyQt5 for final demonstrations
