Deep in systems engineering and AI architecture mode. Most developers live comfortably within layers of abstractions—I prefer to tear them down. Exploring the extremes of the software stack, from custom database engines to open-weight reasoning models.

🧬 Metanthropic Lab – Independent AI research org tackling model alignment, reasoning, and bare-metal efficiency.

• 🧠 Arvi 20B - Open-weight MoE reasoning model matching frontier performance with only 3.6B active parameters.
• ⚡ TealBase - Extreme concurrency Backend-as-a-Service (BaaS) built on Go and PostgreSQL with sub-50ms CDC pipelines.
• 🔌 httpfromtcp - An HTTP/1.1 server built entirely over raw TCP sockets in Go—manual parsing, zero frameworks.
• 🗄️ VostSQL - Custom object-relational database engine built from scratch in C/C++.
• 🔔 pg_listen - Event-driven C library leveraging libpq for async PostgreSQL NOTIFY signals.
• 💾 SIM8086 - Modern cross-platform 8086 architecture emulator and assembler.
• 🎥 Chromabase - The "infinite storage glitch"—encoding and storing files inside video formats via Rust.
• 🎯 M-NAAR Specification - Neural Ablation via Attention Refraction to execute robust knowledge deletion and eliminate hallucinations.

• Mastering the metal - Writing custom database engines, TCP servers, and event-driven C libraries.
• Pushing the AI frontier - Training MoE routing models and developing deterministic neural ablation protocols at Metanthropic.
• Scaling infrastructure - Bypassing standard HTTP polling with WebSockets and deep PostgreSQL RLS integrations.
• Writing about systems internals - Documenting my architecture deep-dives and research on ekjot.me.
• Pursuing B.S. in Computer Science - Vellore Institute of Technology (VIT).

• Arvi 20B: Democratizing Reasoning with Efficient MoEs
• Solving the Unlearning Trilemma with M-NAAR
• VostSQL: Designing a Database Engine Architecture
• Building an HTTP Server from Raw TCP Sockets
• Fragility of Guardrails: Cognitive Jamming

• Founder & CEO at Metanthropic Lab
• Lead Developer for TealBase

"To build robust artificial intelligence, you must first master the hardware and infrastructure it runs on." I build tools to solve the bottlenecks I encounter, from the silicon level up to frontier reasoning models.