imp3: Interactive tool for image pre-processing and automated pipeline creation

Dataprep: Interative Data Visualisation and Processing for ML insgestion

As a Tech. Lead I have been playing many different roles and responsibilities.

1. Building AI-powered Drug screening (Drug discovery) platform for therapeutic cyclic peptides. This consists of:

   1. • • • Large-scale Alphfold (both Monomer & Multimer) inference and automated structure analysis with vertex AI pipelines.

          • Alphafold deep feature extraction for downstream AI applications

          • Virtual screening of peptides (calculating binding strength) using Molecular docking protocol ADCP. Analysis workflow to measure the relevance and quality of output conformations generated.

          • Molecular dynamic simulation. Workflow for trajectory analysis and Free Energy Calculations.

          • Both Docking and MD with corresponding analysis are done at massive levels with parallel batching strategies over massing clusters consisting of GPU-powered nodes, using SLURM.

2. Leading following delivery projects to success

• Gen Ai chat agent (RAG Powered) on domain-specific complex text documents

• Feature extraction and protein structure prediction at large scale using Alphafold to determine the pathogenicity of mutations

3. I have also been actively involved in scoping and reviewing many upcoming engagements.

   1. • • Designing solutions to challenging problems.   

        • Creating detailed project plans, risk assessments, and mitigation pathways.

4. Recruitment and Training

   1. • • I have been part of numerous senior-level recruitment campaigns   

        • Training initiatives, both within and outside our organizations 

5. Mentoring new talent

I have worked in various capacities on various demanding projects at Quantiphi

1. ML capability development

• Advanced Speech to text with translation

• Extract Medical Knowledge from Unstructured Data

• Celebrity Detection

• Sound Event Detection

• Wrapper info Extraction and Validation

• Sentiment analysis from G form data

• Text to Image Retrieval

2. Delivery Projects

• GPU to TPU migration

• Visual inspection ai 

• Xgboost distributed on multiple GPU nodes with Dask

• Vertex Matching Engine (ScaNN)

• Protein Folding

• Molecular Dynamics Simulation

• Disease prediction using insurance claims data

3. Recruitment

• I have been part of numerous campus interviews

• Training initiatives

Biometric Video Surveillance employing multiple cameras at different places by combining (fusion of) Face and Gait recognition for human identification and focusing on the following four problems:

• Scalability to real-time processing

• Occlusion caused by various factors like scene objects, other targets, clothing, bag pack, etc.

• Increased complexity in a crowded area, causing a drastic fall in the efficiency of the current biometric system.

• The biometric Surveillance system may not find a positive match to the captured image/video because of factors like differences in capturing views and different illumination conditions.

Proposed Area/Field of research –

Biometric video surveillance, Machine Learning, Computer Vision.

I worked as a production support executive at an enterprise data warehouse (EDW). I was responsible for successful daily/weekend/month-end/year-end batch processing.

Technology - Mainframe, Ab-Initio, COBOL, DB2, CICS, Unix/Linux

At Bharat Forge, I worked in the Forge Modernization Division (FMD). I learned a lot about the overall forging process, but my interest, in particular, was industrial automation and robotics. I learned about ROBOGUIDE simulation software that provides offline process programming for FANUC Robots.

I designed and implemented a system that can effectively and efficiently classify incoming emails (based on their contextual meaning) and route the email to an appropriate team that can take action on that email.

The system could automatically scale up or scale down in accordance with the particular needs of different organizations.

• Various classification algorithms were tested and compared to find the best suitable classifier for the text classification problem.

• Various feature construction and feature representation techniques for text classification were tested and compared.

• The effect of different preprocessing steps is explored through experimentation.

• The effect of term interaction along with dimensionality reduction is also analyzed.

From the insight gained through extensive experimentation classification system was optimized for real-time processing. This optimized system was interfaced with Gmail to demonstrate its effectiveness.

This project consisted of three parts:

• Mechanical design

• Pneumatic System design

• Control Unit

The focus was to design an end-to-end machine that can automatically take the input material, weld it, and then cut the product of the required size.

• • • Implementation of Back-Propagation Algorithm for classification (Dataset used Iris Dataset)

    • Interfacing Accelerometer and Magnetometer to Firebird V Robot platform and displaying output on a Desktop PC.

    • Intermediate learning experience with tools: Weka, Orange, Fuzzy Toolbox in Matlab.

    • Presented seminar on advanced propulsion system (Electric propulsion).

    • Analyzed compressible flow through a nozzle using CFD.

    • Obtained satisfactory results explaining the reason for the increase in cross-section area at the outlet to achieve Mach number greater than one.

DRISHTI is a non-invasive biometric surveillance system that captures the face, expression, age, gender, gait, and activities of individuals (target persons or criminals) as well as possible weapons present in a distributed CCTV camera system. 

Information captured is compiled and stored in a database.

DRISHTI secured 1 st prize in Smart India Hackathon 2020. (Sudhir as Team Leader)