CV

Principal Architect AI/ML

Byteridge

Lead the development and execution of AI/ML strategies aligned with organizational objectives. Collaborate with product management, operations and sales to align initiatives with customer expectations. Mentor and develop team members, fostering a culture of continuous learning within the team. Communicate with non-technical stakeholders, influencing decision-making at the executive level.

Machine Learning Consultant

Dataworkz

Created state-of-the-art RAG Evaluation framework using LLM-as-a-Judge. Development of RAG-based QA system using fine-tuned finance embedding models. Web-based Text-To-SQL system enabling natural language queries to retrieve data from structured database.

Graduate Teaching Assistant

Colorado State University

Conducted classes on mathematical topics essential for understanding core concepts in engineering. Responsible for conducting the course end to end, creating content, teaching and grading assignments.

• Teaching a braod range of topics to Sophomores and Juniors.

Graduate Research Assistant

Colorado State University

Developed and published a paper on novel neural network pruning technique, achieving significant compression with minimal accuracy loss. Conducted research in UAV target tracking using POMDP, contributing to advancements in autonomous systems. Served as an appointed tutor, teaching advanced mathematical concepts to engineering students.

• Publication of a paper.

Course Assistant

Colorado State University

Grader for graduate courses, ECE514-Application of Random Processes and ECE520-Optimization Methods.

Senior Software Engineer

Nokia

Led cryptography projects, including end-to-end development of OCSP and authentication frameworks. Directed defect management and mentoring initiatives, ensuring high-quality deliverables and knowledge transfer within teams.

• Award for project management excellence (2008), Critical project delivery (2007)

Principal Software Engineer

Network Security Solutions

Development of Xecure Message Service (end-to-end encrypted messaging servive). Development of secure LAN framework using proprietary parallel port communication.

• Letter of commendation for delivery of flagship product (2005)

PhD

Colorado State University

Electrical Engineering

Master of Science

Colorado State University

Electrical Engineering

Bachelor of Engineering

Army Institute Of Technology

Computer Science

Project management excellence

Symbian Software Pvt. Ltd.

Critical project delivery

Symbian Software Pvt. Ltd.

Letter of commendation

Network Security Solution Pvt. Ltd.

Development and delivery of the flagship product.

Activation-Based Pruning of Neural Networks

MDPI

We present a novel technique for pruning called activation-based pruning to effectively prune fully connected feedforward neural networks for multi-object classification.

RAG-Based QA System

Overview: Developed a RAG based QA system for pdf documents. Technology: Multi-vector retrieval, top-k similarity search, bi- and cross-encoder retrieval. Implemented in Python using LangChain. Embedding vector used was Chroma.

• Generative AI
• Retrieval Augmented Generation

RAG Evaluation Framework

Overview: Evaluation method using LLM-as-a-Judge that provides a measurable metric. Technology: Implemented in Python using Prompt Engineering based on the concept of LLM-as-a-Judge. Evaluated performance against BLEU, ROUGE and BERTScore. Benchmarked results against public datasets.

• Generative AI
• Evaluation Framework
• LLM-As-A-Judge

Text-To-SQL System

Overview: Web-based product enabling natural language queries to retrieve information from an SQL database. Technology: Python, Llamaindex, Flask, HTML, Pandas, Snowflake and OpenAI. Innovations: Plot Generation: LLM-based generation of plots. Sanity Checks: Conflict resolution for values referencing multiple columns or datasets. Dynamic Glossary: Used RAG to include domain-specific vernacular.

• Generative AI
• Text-2-SQL

Neural Network Pruning

Overview: Developed a novel method for pruning neural networks. Technology: TensorFlow using custom callbacks and a custom model. Scikit-learn, pandas, matplotlib and Keras. Achievements: Published a first author research paper and achieved 70-80% network compression.

• Network Pruning
• Feedforward Network
• Weighted Nuclear Norm Minimization

Anomaly Detection

Overview: Applied advanced techniques for detecting anomalies in time-series and categorical data. Techniques: Some of the methods applied were; Isolation Forest, ARIMA, Logistic Regression, K-Nearest Neighbor, Support Vector and Decision Tree Classifier.

• Time series techniques

Target tracking of a Random Walk using Kalman Filtering

Carried out an implementation of the Kalman Filter to track a target moving randomly in two dimension.

• Kalman Filtering

Isolated word recognition using Hidden Markov Model

Isolated speech recognition system using Hidden Markov Model (HMM). HMM is a probabilistic sequence model which are based on Markov chains. The problem can be broken into real world signals in terms of signal models. Solving the problem is a 3 step process in HMMs; Likelihood: Allows us to choose the model which best matches the observation, Decoding: Attempts to uncover the hidden part of the model and Learning: Attempts to optimize the model parameters so as adjust it for the set of observations relates to a class.

• Hidden Markov Model

Orchestrated Management of Heterogeneous Sensors via POMDP

Implementation of Orchestrated Management of Heterogeneous Sensors using Partially Observable Markov Decision Process (POMDP). The problem was formulated as a POMDP ( states, actions, state transition law, observation law and cost function). Features identified and incorporated into the POMDP framework, such as feedback from humans (intelligence assets) on the targets and influence of the intelligence assets feedback to the path of UAVs. Tuning of weights to incorporate feedback from intelligence assets was based on augmented Lagrangian method applied on constrained MDPs.

• POMDP

USB based block device driver for removable media

Developed a USB based block device driver for a SD card hosted on a LPC2378 board running an ARM7TDMI-S 32 bit RISC micro-controller.

• Device Driver

Plugin for Hardware-based crypto operations

Development of security token framework, to support hardware-based cryptography operations. A sample plugin was developed.

• Cryptography

Authentication Framework

Development of password based authentication framework. Using it to enable per user access to the device. Development of a tool to seamlessly migrate existing device data.

• Cryptography
• Identity Access Management

Development of Online Certificate Status Protocol (OCSP)

Implementation of the protocol was based on RFC 2560. It was used for certificate validation during application installation in Symbian based phones.

• Cryptography
• OCSP

Secure SMS (XMS - Xecure Message Service)

Development of server side components to process registration and activation for enabling peer-to-peer security for mobiles.

• Cryptography

Secure LAN framework

Development of communication protocol between server and middle-ware for secure communication. Server was isolated through protocol running on parallel port.

• Proprietary Communication Protocol