My Projects
A collection of projects showcasing my expertise in embedded systems, backend development, and machine learning.
Omni-Channel Voice Agent (Community Edition)
An ultra-low latency, real-time omni-channel voice platform featuring dual AI engines: Local CUDA Stack (Whisper.cpp + Clef-Flash + Qwen3-4B + Kokoro TTS) and Cloud Indic/HD Stack (Sarvam AI + ElevenLabs Turbo v2.5). Built on Rust (Axum async carrier + AVX-512 SIMD audio DSP), in-browser Silero Neural VAD v5 (WebAssembly), sub-microsecond in-memory clinical tool execution (<1µs), and node-based visual workflow studio designer.
All Projects
QuantStream: Real-Time Blockchain Mempool Telemetry & EVM Block Builder Engine
QuantStream is a production-grade, distributed real-time blockchain telemetry and block execution (matching) engine built entirely in Rust (with a Python PyTorch training pipeline). It demonstrates the implementation of the Lambda Architecture—optimizing a high-frequency 'Hot Path' for sub-millisecond AI inference while securing a 'Cold Path' for durable event persistence and tiered multi-database analysis.
Custom CUDA & Triton GPU Kernels
Hand-optimized low-level CUDA C++ and Triton JIT kernels for FlashAttention, PagedAttention, KV cache packing, and non-autoregressive decision model inference.
Zero-Allocation Go Microservices
Ultra-high concurrency web service engine built on Go/Fiber and fasthttp with memory buffer pooling, lock-free SPMC channels, and SIMD JSON acceleration.
Resilient IoT Ingestion Engine on 2-Core Cloud
High-throughput asynchronous MQTT telemetry pipeline with memory-bounded queuing and time-partitioned database micro-batching on ARM64.
Limit Order Book Matching Engine
Deterministic price-time priority (FIFO) matching engine with stochastic jump-diffusion modeling and low-latency order execution.