Projects

Case studies, open-source systems, and deep-learning infrastructure.

AI & ML2026

Self-Attention Kernel & Inference Engine from Scratch

Implementation of FlashAttention-style tiled multi-head attention kernels in C++/CUDA and custom PyTorch C++ extensions with benchmark comparisons.

Speedup vs Naive PyTorch3.8x
Peak SRAM Utilization91%
C++20CUDAPyTorch+2
Data Engineering2026

Distributed Low-Latency Stream Processor in Rust

A zero-copy, fault-tolerant distributed streaming engine built in Rust for processing 1.5M+ events/sec with sub-millisecond p99 stateful window aggregations.

Throughput (Single Node)1.48M ev/s
p99 Window Latency0.85 ms
RustApache ArrowKafka+3
GenAI2026

Multi-Agent GenAI & RAG Evaluation Harness

An automated regression and evaluation suite for testing complex RAG retrieval, multi-hop reasoning, tool invocations, and semantic drift with synthetic dataset generation.

Eval Throughput10k runs/hr
Judge-Human Correlation0.91 Pearson
PythonPyTorchvLLM+4
Systems2025

Vectorized Columnar Execution Engine for Parquet Lakehouses

A lightweight vectorized OLAP query engine in C++ executing SQL aggregation, projection pushdowns, and dictionary-encoded string filtering directly over Parquet files.

Scan Speed (SSD)4.2 GB/s
Aggregation Speedup6.4x
C++20Apache ArrowParquet+2
Data Engineering2024

Automated Flight Data Pipeline with Medallion Architecture

A robust end-to-end data pipeline using Apache Airflow, Apache Spark, PostgreSQL, and AWS S3 with Medallion architecture (Bronze, Silver, Gold).

Pipeline ArchitectureMedallion (Bronze/ Silver / Gold)
Transformation EngineApache Spark
Apache AirflowApache SparkPostgreSQL+3
GenAI2024

MindScope: Modular LLM Persona-Aligned Insights & EDA Engine

A modular Python package leveraging Large Language Models for persona-aligned automated exploratory data analysis, summarization, and goal generation.

Deployable Personas10+ Profiles
ModulesSummarizer, EDA, Goals
PythonOpenAI SDKLangChain+3