I am a cross-domain systems architect. I build prototypes and proofs-of-concept that solve hard bottlenecks, and I use AI as an execution engine to build them fast.

I fundamentally reject the industry's default brute-force paradigm of throwing more compute at a problem (mostly because GPUs are too expensive). My philosophy is simple: understand the physical and mathematical constraints first, prototype the solution in days, and let telemetry win the arguments.

I don't build the polished 1.0 product. If you're looking for CSS animations, I'm the wrong guy. I build the foundational architecture that makes the product possible. When I don't have a specific problem to solve, I run on pure curiosity—usually looking for exactly where systems break under pressure.

• ASH-KV (Work in Progress) Tiered KV cache management for LLM inference. 8-number policy surface, fault-tolerant migration engine, hardware-agnostic codec layer.
• InferBench A benchmark built specifically to stress-test LLM serving systems under concurrent load. The signature "cliff finder" pinpoints exactly when and why an API will inevitably catch fire in production.

• Fluidic Equilibrium (Work in Progress) Fighting the quadratic attention bottleneck. A hybrid SSM architecture (Liquid-S4 + CfC + DEQ) implemented completely from scratch just to see how far state space models can really go.
• CKDG Cyber-physical attack detection running on air-gapped local LLMs. Built on the thesis that statistical anomaly detection is just guessing with extra math, whereas physics provides deterministic constraints.
• rocm-doctor A zero-dependency diagnostic tool for AMD ROCm installations. Because diagnosing GPU drivers shouldn't require a guessing game and a blood sacrifice.

• Contract-First Design: Code without strict boundaries is just technical debt waiting to happen.
• Trust, but Verify: I don't trust assumptions about hardware, drivers, or model outputs until the telemetry proves it.
• Rapid Prototyping: I direct the AI to write the boilerplate so I can spend my time designing the architecture and profiling the bottlenecks.

📫 Reach out: ijasdatai12@gmail.com | Always open to talking about hard architectural constraints, designing fault-tolerant systems, or what happens when you push hardware past its limits.