Building AI Agents for the Future Using Advanced-Level Flow Engineering.

• Solving data as a service in a privacy preserving way • Finalist (Top 4 %ile) for Build for Bharat Next Gen Venture organised by ONDC and Antler• Built a deep tech infrastructure to solve the data sharing problem in privacy preserving way for ONDC and did POC on 5 Clients but need to drop because it was too early for ONDC itself being in early PMF stage

Studied $500B Credit gap for MSME Lending using Indian Stack and experimented with cash flow based underwriting model

NAUKRI CLIENT GROWTH - B2B• Initiated and established this charter, implementing a proper framework for CRM & Naukri transaction flow, which had been decentralized and untapped for a long period.• Developed AI-driven models to optimize corporate client reactivation & churn prevention, while designing a collaborative architecture with Naukri & AmbitionBox to generate new client leads through statistical signals.• Facilitated cross-functional collaboration by leading meetings with senior stakeholders, including heads of Sales, B2B GTM, and ERP, to execute on ground.• Unlocked ₹90-220 Cr revenue opportunity through client reactivation, with ₹22 Cr already achieved in FY23.• Gained an edge with early churn signalling with precision of 72% and further enhanced the sales team’s negotiation skills and client understanding through language-generated reasoning.NAUKRI FAST FORWARD - B2C PREMIUM SERVICES• Single-handedly led and revamped the data science backing Naukri Fast Forward Business, achieving highest YOY revenue growth of 26% in the last 5 years under my tenure.• Built advanced ML models for two major projects, resulting in improving monthly lead conversion from 8.1% to 11.8% and monthly apply rate from 2.5% to 5.2% in case of recommendation engine.• Streamlined heavy operation driven tasks through machine learning, leading to a notable increase in recruiter engagement from 65% to 77% for paid users.• Developed AImodel for organic customer growth, replacing heuristic logic and resulting in improved promotional discounts monthly conversion from 0.9% to 1.5%.

Facial Expression Recognition Improvement• Worked on improving real-time feedback on facial expressions during a mock interview.• Analyzed the various classes of the data to combine and reduce them to expression classes needed according to the client.• Investigated various papers based on FER and developed modules like kernelized dense layer, self-attention, and integrated with the existing model.• Thorough hyper-parameters tuning ablation study due of the limitation of model size as per the product demand.• Inferences: Increased overall accuracy by 4 percent and increased class-wise precision and recall by a good margin for the existing base model.Appearance Recognition & Background Classification• Analyzed the appearance and background of the interviewee during a video interview.• Modelled various elements such as professional attires and background conditions to improve thefeedback system.• Analyzed the video frames using Dlib and OpenCV to extract relevant feature elements for the different cases.• Designed and developed tagging panels for storing and generating input data to train machine-learningmodels.• Developed a deep learning-based algorithm for the classification tasks using TensorFlow 2 and 1 both.• Error handling and resolving library-level conflicts while systematic TensorFlow 2 to 1 migration.ML Modular Library• Designed an ML module structure to standardize the general development process for any product.• Developed heuristics to make the development of new projects efficient from the existing projects.• Converted the whole pipeline system to a modular form that is easy to use.• Modelled a singular app-like interface that can directly be used while developing the ml-based backend of new products.

Traffic Flow Prediction Using Deep Learning• To design & build a multi-agent dynamics system that will help streamline traffic to avoid jamitons.• Data: Time series traffic density counts for various roadway sites as collected by various sensors.• Implemented Seq2Seq LSTM as an MVP.• Did investigation on related domain research papers.• Developed a new LSTM architecture using a pooling module.• Inferences: Reduction in RMSE value by 12 % for sensors experimented, retrieved related sensors.

Antaragni - the annual cultural festival of IIT Kanpur - is one of the largest and most anticipated festivals throughout Asia• It involved creating banners and posters for the various events planned by different festival cells.• Developed design strategies for sponsors' mass reach and awareness• Used tools like Adobe Photoshop CC and Adobe Illustrator at a professional level for design purposes