benzo
I build products at the intersection of AI, security, and developer tooling. I enjoy creating fast software, solving difficult engineering problems, and competing in cybersecurity competitions.
Hi, I am Tushar 👋
I focus on building and breaking complex systems. As a Software Engineer, I design robust architectures ranging from minimal, highly-responsive user interfaces to scalable backends. My primary AI research and engineering efforts are concentrated around AI Systems, LLM Agents, RAG pipelines, and Multi-Agent Orchestrations.
On the security front, I research Web App Security, Cloud Security, and offensive operations. Balancing full-stack engineering and security audits allows me to construct applications that are high-fidelity, high-performance, and resilient against modern attack layers.
Current Focus
Building AI Developer Tools
Developing next-generation code analysis utilities and static checkers integrated with reasoning models.
Researching Agentic Systems
Structuring safe state machines and routing graphs for multi-agent loops to eliminate looping traps.
Cybersecurity Research
Auditing application dependencies and analyzing novel vector injection paths inside distributed networks.
Competing in CTFs
Actively solving high-difficulty challenges and engineering attack/defense targets for competitive arenas.
Indie Product Dev
Shipping minimal, high-utility tools addressing developer problems with quick iteration cycles.
Technical Writing
Documenting deep-dives on systems architecture, cybersecurity remediations, and performance logs.
Featured Projects
EHAX CTF Platform
activeA highly resilient, containerized cybersecurity capture-the-flag platform designed to handle thousands of concurrent players.
Latest Writing
Suricata: A Complete Guide to Modern Network Threat Detection
A comprehensive guide to Suricata, its architecture, rules, deployment modes, and integration with modern SIEM platforms.
vLLM: A Modern Inference Engine for Large Language Models
An exploration of how vLLM serving is optimized using Continuous Batching and PagedAttention.
Caching: A Complete Guide to Cache Design, Policies, and Eviction Algorithms
A comprehensive guide to understanding caching, why it exists, how it works, cache eviction algorithms like LRU, LFU, FIFO, ARC, TinyLFU, and how modern systems use them.
Understanding Model Parameters: Total Parameters vs Active Parameters
An explanation of how parameters are used during LLM inference, and the key differences between total and active parameters in Dense vs Mixture of Experts (MoE) architectures.