Graphene

social network graph language interpreter

Graphene makes graph manipulation and social network data analysis convenient using a purpose-built interpreted programming language. The motivation for our language is the massive commonplace use of graphs and graph based data mining algorithms in today's software world. At the same time, we see a large bubble of social network and social network-like applications which manage a large backend of data which can usually be represented using a graph structure.

Most of today's languages do not provide out-of-the-box or easy to use features for graph initialization, operations and management. Graphene will provide this interface to be able to support generic graph algorithms as well as specific social network applications based computations on graph-like data structures.

Code can be found here.

For a live demo, go here.

CreditLens

Built for Mortgage Portfolio Management and Intelligent Account Handling

RESTful, powered by RoR (Ruby on Rails), running on AWS.

See it running here.

CloudVisor

Designed to automate monitoring of VMs and harden security. Agent based.

Control Flow Integrity

Kernel Hardening to subvert control hijack attacks such as ROP.

Multiprocessor Network Subsystem Performance Tuning

sk_buff manipulation to force packet processing by specific cores to ensure lesser core cache miss/hit.

Modified Completely Fair Scheduler

Optimize power consumption through scheduler decisions when battery power reserve goes below threshold. Non critical applications are forced to free network access priviliges.

Distributed Intrusion Detection System

Project majoring in Artificial Intelligence & Network Security, to research and develop the next generation of network systems capable of defending the network by evolving. Genetic algorithms such as clonal selection used in a distributed non-cooperative system to generate rules. Activities were mapped, data mined and anomalies were identified using clustering algorithms and the Hamming distance, Euclidean distance and Overlap area methods.

Rules function on multiple layers of OSI Layer and enact the required action such as event flagging and firewalling. While neural networks are more effective at training, genetic algorithms have more effect at learning. Learning could not be done on-the-fly. There had to be cells that evaluate the threshold values and then reflect it system-wide. We maintained a hierarchical architecture for the detectors with the primary at the gateway performing passive network captures so as to be more effective.

Your very own File System

Distributed file system, built on top of the VFS interface, with distributed locking, crash recovery and paxos consensus implementation.

Secure Object Stores

Distributed Fault-tolerant Lock Server