Projects

Built a concurrent HTTP client library in C using direct Linux system calls (no libcurl/libc networking), implementing TCP sockets, HTTP/1.1 request/response handling, and buffer-based parsing. Extended the library with a custom thread pool and thread-safe task queue built from scratch using pthreads, mutexes, and condition variables, enabling async HTTP requests with a callback API. Achieved a 3.8x speedup over sequential execution across 10 concurrent requests.

Built a parallel file encryptor/decryptor in C++ implementing four concurrency architectures - thread pool, multiprocessing, hybrid, and async I/O (io_uring) - empirically identifying disk I/O as the bottleneck via instrumentation and perf stat, and demonstrating that async IO achieves equivalent throughput with 7x lower CPU consumption than multiprocessing on a 6-core machine.

Built and deployed a multimodal PyTorch model for video emotion analysis using text, video, and audio encoders with late fusion. Deployed on AWS SageMaker with a production Next.js frontend featuring S3 direct uploads, API inference, and quota enforcement, implementing a complete ML-to-production pipeline.

Built a complete ping utility in C across two implementations. Part 1 uses raw sockets and manual ICMP packet construction, checksum calculation, RTT measurement, and signal handling. Part 2 is an ongoing lower-level rebuild manually constructing every network layer - ICMP and IP headers byte by byte using a two-struct serialization pattern, with Ethernet (Layer 2) in progress. Verified output against tcpdump and production iputils source. No networking libraries used.