Regulatory and COMPliance AGENTIC CHECK

Built a multi-agent compliance layer auditing data residency, PII handling, tool-permission scope, and model output risk across all portfolio AI applications against compliance frameworks.

CPX-CSM-Agent (Customer Service Management Agent)

Built a LangGraph-based agentic pipeline that automates client email triage, intelligent response generation using hybrid RAG (BM25 + semantic search), and end-to-end ticket lifecycle management on the CPX NOVA Portal. Features human-in-the-loop manager approval via email/Slack/Teams, priority-based SLA routing, and automated client notifications.

OT/ICS Exposure Intelligence Platform

Designed and built a full-stack AI powered OT/ICS cyber exposure intelligence platform targeting UAE/GCC critical infrastructure. Developed a Shodan-based data ingestion pipeline scanning 15+ OT/ICS protocol categories (Modbus, BACnet, DNP3, EtherNet/IP, Siemens S7, MQTT, SCADA/HMI, etc.) with automated extraction into PostgreSQL. Built a FastAPI REST backend serving a React dashboard featuring an interactive geospatial exposure map, risk scoring, category-wise analytics, trend visualization, and a paginated data explorer. Tech stack: Claude, Shodan API, PostgreSQL, FastAPI, React, Docker.

Intelligence Threat Detection

Developed a comprehensive threat intelligence platform for proactive security monitoring, featuring detection modules for rare user-agent analysis, phishing identification using ML models, suspicious parent-child process relationships using unsupervised learning, anomalous VPN access patterns, unauthorized remote logins, and credential compromise detection.

Entity Resolution using Large Language Models(Gemini 1.5 Pro)

The solution was deveoped for a client in cricket domain. The query asked by the users had incorrect names, spelling mistakes, nicknames but the database had proper names of the cricketers and it was necessary to identify Player Name, Venue, Tournament, Year etc. So Named Entity Resolution was done using LLM and desired results were obtained as per client expectations.

Deployment of Instance Segmentation Model on GCP using Vertex AI

The clients existing ML models were migrated to Vertex AI on Google Cloud Platform (GCP) and orchestrated ML workflows in the cloud. Secure storage of captured video data on GCP and transformation of the current UI into a cloud-hosted web application were undertaken. Combination of GCP services along with orchestation in Kubeflow was done.

Segment Anything Model(SAM) in Medical Imaging

The Segment Anything Model (SAM) is a state-of-the-art image segmentation model that was introduced by Meta. The SAM model is designed to be promptable, which means that it can generalize to new image distributions and tasks beyond those seen during training. This capability is achieved through the use of prompt engineering, where hand-crafted text is used to prompt the model to generate a valid response for the task at hand.

Model Training and Deployment using KubeFlow

The Kubeflow project is dedicated to making deployments of machine learning (ML) workflows on Kubernetes simple, portable and scalable. My goal is not to recreate other services, but to provide a straightforward way to deploy best-of-breed open-source systems for ML to diverse infrastructures.

Medical Document Classification in BERT

The purpose of this project was to classify and extract data from medical documents shared by users. For classification of document, a model in BERT was developed and rest of the data was extracted by AWS Textract and AWS Medical Comprehend using Named Entity Recognition(NER).

BERT stands for Bidirectional Encoder Representations from Transformers. It is designed to pre-train deep bidirectional representations from unlabeled text by jointly conditioning on both left and right context. The data used in this project is the Medical-Abstracts-TC-Corpus dataset. The dataset has 5 categories - Neoplasms, Digestive, Cardiovascular, Nervous and General pathological conditions. The dataset had 14438 annotated samples.

Analysis and study of Particulate Matter levels through a network of low cost IoT devices in IIT Delhi campus

The project has been funded by IIT Delhi under Smart Campus Initiative. It aims to identify and study the variations in Particulate Matter at various locations, with respect to height, predictions and hotspots over an entire area.

This spatial map shows the levels of PM at various locations with yellow being low values and red having highest values.

The PM levels show a perodic trend with values shifting from low to high in day night period.

NASA FIRMS data provides real time active fire data from its two satelites Aqua and Terra. This data provides the crop burning data which can be compared with the PM 2.5 as crop burning was one of the major reasons for rise in PM levels during winter months.

Secondly, Effect of the Odd-Even Government policy on the the air quality.

Comparison of PM data with Weather parameters

Forecasting and Predictions

Regression

Time Series Forcasting - SARIMAX

Air Quality Accessment and Monitoring in the region of Delhi using a cluster of 200 DIMTS Buses

The project incooperates installation of low cost sensor devices through a partnership with Delhi Integrated Multi-Modal Transit System(DIMTS) in 200 buses. Each bus makes 16 trips daily covering different routes each 30-40 kms. In the first phase we have installed 10 devices in various buses. The project aims at collecting particulate matter(PM) across different terrains, weather and times of a day.

The analysis will provide remarkable insights into the different factors affecting PM. These factors will further be ranked to suggest remedial actions and devise appropriate PM control policies.

The device has been developed at IIT Delhi on our own designed PCB and then routed to single board computers i.e RaspberryPi followed by mounting them in the bus.

Data Pipeline

The data coming from these bus devices is send over the internet to the Google Cloud Platform(GCP) where it undergoes data pipeline(Data Storage, Pub/Sub, Dataflow) and finally is stored in the Big Query.

Heat Map

Classify satellite data of the region of Delhi into various zones and then correlate with Particulate Matter levels.

This project aims at identifying various regions of the state of Delhi into Industrial, Comercial, Residential, Water Bodies and Barren Areas. This classification has been done using Masked-RCN from geo-spatial satelite data of Delhi.

This classification will be co-related with the PM 2.5 Mobile Bus data to observe the tpe of area in hotspots.

Body Scanner Image Management(BSIM)

A Data Logging Computer for Airport Authority of India to store data from various devices present in the airports. These devices comprises of Full Body Scanners, Shoe Scanners, Boarding Pass Bar code Readers. The data is stored locally as well as stored in the central database.

https://economictimes.indiatimes.com/industry/transportation/airlines-/-aviation/84-airports-to-install-body-scanners-by-march-2020/articleshow/69619539.cms?from=mdr

Extensive Image Processing algorithms ranging from feature matching, template matching, compare histograms to color detection and character recognition were tested to achieve results in least possible time.

Automatic dependent surveillance(ADS-B)

A surveillance technology in which an aircraft determines its position via satellite navigation and periodically broadcasts it, enabling it to be tracked. The information can be received by air traffic control ground stations as a replacement for secondary surveillance radar, as no interrogation signal is needed from the ground. It can also be received by other aircraft to provide situational awareness and allow self-separation.

My goal here was to decode the raw encrypted data coming to the receiver and then convert it into readable format for air traffic control.

Hardware and Design of UAVs

Worked with BotLab Dynamics(IIT Delhi) in hardware and design of UAVs for various competitions.

Google Sites

Report abuse