Meet me
Hello!! World, This is Shashank, a 21st Century Data Science Specialist striving assiduously to augment competencies and innovate ways for human - machine bilateral interaction. Overcoming multifarious challenges posed by seemingly impregnable barriers across all channels of natural communication. Focusing avidly on image, speech and natural language processing that leverages Generative AI driven Large Language Models (LLMs).
Over past 9 years, I have achieved a competitive knack to specialise in designing custom APIs and Application Softwares as per AI industry standards and regulations. Apparently soon after graduating from engineering school, I embarked in a crusade to discern and disseminate AI to its last atom by harnessing the following skills in my arsenal :
Transformers · Kubernetes · BERT · spaCy · TensorFlow · Exploratory Data Analysis · Predictive Modeling · Microservices · Optimization Algorithms · Algorithms · Microsoft Azure · Amazon Web Services (AWS) · Pattern Recognition · Cloud Services · Analytical Skills · Data Pipelines · Docker Products · Big Data · Python (Programming Language) · Natural Language Processing (NLP) · Machine Learning · Deep Learning · Computer Vision · Keras · SQL · Robotics
I live by a code that reciprocates with progressive work-style and retrospective improvement paradigm.
Explore the meticulously designed portfolio, which encompasses commissioned work, client projects, community contributions and an archived repository of interesting surprises.
Always excited to meet and greet leaders, visionary and entrepreneurs to stay motivated and get inspired. Don’t hesitate to spam my inbox if you’d like to learn more about me and my work-life. It will make me rejoice.
Welcome to Shashank's Webspace.
"Past Recognitions - Present Engagements - Future Endeavours"
Leveraging the latest AI breakthrough in Generative AI and operating on ;
Foundational LLMs with commercialised and open-source licenses (GPT-3.5-Turbo, ChatGPT, Flan-T5-xxl, BERT-large, GPT4All, LLaMA, UL2 and Cohere). Fine-tune to custom domain specific dataset and impose control and moderation techniques to limit hallucinations in generated contents for Legal, Tax, Audit and consulting inquiries. Production load the fine-tuned LLM models with all its trained weights and state parameters to conduct real-time and batch inferencing on customer data in a secured environment with CPU/GPU compute using edge-devices.
Designed and delivered APIs and AI software products featuring LLMs to various clients from retail, automotive, finance, pharmaceutical, sales, e-advertising, robotics, e-learning and digital streaming sectors. Have cognitive outlook towards convoluted analytical engine behind all forms of data generated continuously that can be leveraged to bring ROI with quick market turn-around time. Strong believer of symbiotic working culture with clients right from beginning.
In future, I would like to extend and pursue project endeavours in the direction of Artificial General Intelligence (AGI) for the benefit of humanity.
Project Highlights
Evidence backed extractive Question Answering with fine-tuned LLM prompts through vector databases.
Framework : Transformers, Hugging Face, Spacy, NLTK, OpenAI
LLMs : Flan-T5 XL (Google), Faln-T5 XXL, GPT4, ChatGPT, BERT, LLaMA, BLOOM
Database : PostgreSQL
Developed a knowledge processing pipeline for exclusive legal domain documents to get processed and stored as vector databases. It gets linked with prompt based Question Answering system with LLM as backend that interprets user instructions and selects the relevant documents vector from database and synthesises most appropriate and optimised response to render into cross-platform mobile application.
Machine Learning prediction to determine
conversion matrix for identifying the probability of a potential customer to migrate from free to paid version of Microsoft Subscription based services for APAC zone.
Framework: Apache Airflow, Azure Databricks, Flame graphs
Libraries: Statistics, NumPy, SciPy, XGBoost, Decision-Tree, Random Forest, Pandas
Database: MySQL Workbench
Containers: Dockers for Windows
IDE: Spyder, PyCharm
Advanced ML driven XGBoost model is used to learn and discern the features associated with numerous customer centric campaigns and programs hosted by Microsoft from time to time. Further it underwent pre-processing stage to handle missing, noisy and inconsistent data to make them clean and model ready. Various boosting and efficient learning algorithms are implemented to fit on data and to make it robust enough to perform generalized predictions on new unseen data set. This execution was tightly coupled with steps of model optimization and hyperparameter tuning to achieve the desired value of evaluation metric.
Face recognition-based Person Detector
system for remote surveillance of facility and logging of detected faces in database
Framework: Django (Python for Web), PyDev, Convolutional Neural Network(CNN)
Database: MySQL Workbench
Libraries/APIs: TensorFlow, Dlib, OpenCV, MTCNN, Encoders, FaceNET
The project incorporates the implementation of Dlib and Tensor flow deep learning library with OpenCV to detect, recognize and log the human faces in database on real-time basis. It was designed to provide biometric based authorization and authentication for access into any restricted enterprise facility.
Face-ID based authentication and Smart Recommendation system for a US-based Airline client to give personalized travel experience to passengers at Airport Lounge
Framework:
Dynamic Web App, Django (Python for Web)
Libraries/APIs:
Microsoft Face APIs (Azure Cognitive Services)
IoT Platforms:
ThingSpeak, ThingWorx, Microsoft Azure IoT
The project comprises of a web-portal designed for Airline passengers to expedite the Check-in process hassle-free at every stage from Airport entry to boarding flight with Face-ID. The system was also integrated with machine learning library to provide promotions and recommendations evaluated from traveler's history.
Autonomous self-driving Golf Cart powered by Computer Vision and Artificial Intelligence
Framework: CUDA, ROS, Gazebo, REST API, YoLo Object Detection
Libraries: TensorFlow, Keras
Hardware: Raspberry Pi, Arduino,
Languages: Python, C, Embedded C, Curl
This was designed as a part of special endeavor to build and transform an existing electric drive of Golf-cart into an autonomous drive. It was developed and implemented on-premise with the incorporation of computer vision, semantic segmentation, object detection and Internet of Things. The objective is to convert the golf cart completely into voice-operated personal cab in which user can issue voice commands to start and stop the ride. The powerful NVIDIA computing systems on-board predicts the steering angle and direction of cart movement based on the trained deep-learning model.
ThingWorx based HMI design for
displaying real-time distance to impending obstacles as viewed from a dash-camera attached to mobile robot
Hardware Processor: Raspberry-Pi Model 3
Libraries/APIs: Image Processing, OpenCV in python
IoT Platforms: ThingWorx
This was a Proof of Concept (PoC) implementation to demonstrate the capability of ThingWorx composer and studio environment to seek real-time obstacle distance form OpenCV based algorithms and cascading it to ThingWorx platform for rich analytics and immersive visualization.
October 2017 - Present
Data Science, Machine Learning & Artificial Intelligence
Proficiently delivered innovative Proof of Concept(PoC)/ Minimal Viable Product (MVP) design and quick implementation with open-source ML and LLM frameworks to tryout the viability and performance. Have exclusive exposure to handle, manage and train Deep Learning Models on unstructured data accounting for vast volumes of images, audios, videos and text in decentralised fashion.
Productionized efficient and custom-built Generative AI applications leveraging LLMs like Google-Flan T5, GPT-3.5-turbo, Sentence BERT, and BLOOM. It includes abstractive QA, extractive QA systems, evidence-extraction apps, chat plug-ins and content-moderation APIs.
Designed, built, trained, validated and deployed highly robust and Business Intelligence driven Machine Learning Algorithm on the cloud space of AWS / Microsoft Azure and drawn intuitive predictive analytics/ insights to realize faster ROI.
August 2015 - October 2017
Machine Learning, Edge computing & Robotics
Designed Machine Learning APIs and Applications driven by computer-vision and speech-recognition algorithms with Django backend on edge computing devices like Raspberry Pi, Arduino and NodeMCU.
Delivered AI software products to clients from retail, automotive, finance, pharmaceutical, sales, e-advertising, robotics, e-learning and digital streaming sectors.
June 2014 - August 2016
Embedded Software Development
Explored and exploited the limits of embedded hardware for rapid-prototyping of standalone and server-based applications. Earned profound expertise in working with Raspberry-Pi, Arduino and NVIDIA chipsets to deploy embedded softwares and write unit-test cases for functional modules.
ThingWorx based HMI designed for rich analytics and immersive visualization. displaying real-time distance to impending obstacles as viewed from a dash-camera attached to mobile robot.
Materialized BLE Beacons based Indoor Navigation System operating within building or enclosed premises for finding
directions to intended destinations on the go using precise location fingerprinting of wireless signatures.
August 2015 - November 2018
Application Software Development
Developed high quality web, mobile, server-based and server-less applications leveraging the advancements in Internet of Things, Node-RED, Machine Learning, Computer Vision and Artificial Intelligence.
The major highlight being Face-ID based authentication and Smart Recommendation system for a US-based Airline client to give personalized travel experience to passengers waiting at Airport Lounge.
All Videos
Next Big Thing - Beacons- What they'll do for retail
Vision AI - Persona Detection
IoT - Smart Warehouse
By
Abraham Lincoln
"The best way to predict the future is to create it"
Let's Connect!
Interested in learning more about my work or how we can collaborate on an upcoming project? Reach out today.
