**Kaggle achievements**:
- World’s first Quadruple Grandmaster
- Top 5 in all categories
- Kaggle profile: http://www.kaggle.com/abhishek

Get my book: Approaching (Almost) Any Machine Learning Problem for FREE: https://github.com/abhishekkrthakur/approachingalmost

I post a lot of cool stuff from time to time on my YouTube channel. Subscribe me on YouTube: https://www.youtube.com/AbhishekThakurAbhi

I hold a Master's Degree in Computer Science from the University of Bonn. In my free time, I like to take part in machine learning competitions and have taken part in over 200 competitions. My research interests include automatic machine learning, deep learning, hyperparameter optimization, computer vision, image analysis and retrieval and pattern recognition.

Abhishek Thakur

Microsoft

Part of Microsoft's Bing Defensive team, where we are spearheading a transformative approach to search by harnessing the power of Large Language Models. My role within this innovative project is as a Responsible AI Specialist, and my primary objective is to guarantee the model's performance in terms of trustworthiness, fairness, and safety.

As a dedicated member of the Bing Chat team, I am tirelessly committed to enhancing the user experience, ensuring that Bing Chat provides the utmost value while steadfastly upholding Microsoft's unwavering commitment to responsible AI practices. By proactively addressing issues related to Responsible AI, I contribute to the development of a chat-powered search experience that is both revolutionary and reliable. I am enthusiastic about the impact our work will have on the way people interact with search technology and look forward to continuing to drive excellence in responsible AI at Microsoft.

Data Scientist

Cymorg Inc.

LLM Integration
 Aimed to shift from a fixed solution-oriented chatbot to a more conversational format, with a focus on natural and dynamic interactions. Substitute the entire NLP pipeline with an LLM capable of comprehending user intent.
 Addressed challenges related to controllability, RAI (model biases), limiting model scope through extensive experimentation and iteration involving prompt engineering
 Set an ambitious objective to investigate open-source instruction-tuned models with fine-tune support, also exploring alternative solutions like AWS Bedrock.

Gibberish Filter | Chatbot Preprocessing
 Gibberish Detector reached a milestone of 975k downloads in a month, for multiclass classification of incoherent sentences. Implemented a comprehensive categorization system for gibberish detection, including categories such as Noise, Word Salad, Mild gibberish, and Clean.
 Jumped to the 3rd page of the Most Downloaded models in the world on HuggingFace, competing with models from organizations like Google, Microsoft, and Meta. Top 10 in the world amongst the models by individual contributors!
 Developed using Huggingface AutoNLP - fine-tuned on an augmented NLU dataset to achieve high accuracy in detecting gibberish.
 Deployed the model using AWS Sagmaker converted directly from there HFHub - providing a user-friendly and accessible solution for integrating the gibberish detector into various applications and systems.
 For more info, please refer https://huggingface.co/madhurjindal/autonlp-Gibberish-Detector-492513457

Serverless Migration
 Developed a hybrid approach that combines the strengths of both provisioned and serverless inference, offering advantages such as eliminating cold starts, achieving infinite scalability, and enhancing flexibility.
 Migrated over six services from a provisioned architecture to a hybrid framework thus notably reducing operation costs, leveraging Docker containers deployed through AWS Serverless Application Model (AWS SAM).

Founding Engineer NLP

Chatbot Encoder Module
 Aimed at optimization of Simulation Chatbot with current word-embedding level model as benchmark 
 Deployed the sentence-embedding level USE model on Docker Container for semantic text similarity
 Devised a selection pool reduction algorithm tuned using an analytical hyperparameter search in collaboration with founding director
 Conducted evaluations of various sentence embedding models within the e2e pipeline- encompassing similarity prediction and subsequent selection pool reduction.

Client-Level Finetune | Chatbot Encoder Pipeline
 Aimed at a client-level finetuned chatbot encoder pipeline for better business contextual understanding
 Final goal to create a training pipeline with input as vanilla encoder pipeline and output as client-tuned pipeline consuming autogenerated contextual simulation data
 Added complexity stemming from non-differential nature of 2nd part of encoder pipeline and the interdependence

Natural Language Processing Intern

-> Received Pre-Placement Offer (PPO)

 Targeted towards improvement of the existing deep contrastive ranker for PAA on Bing Search Engine
 Experimented on relevance & CTR datasets, custom contrastive losses, CTR bias correction techniques implemented using Pytorch
 Successful flight post a full exploratory analysis with big mover & intersectional analysis on past data

Machine Learning Intern

Goldman Sachs

 Analysed a password database breach & proposed possible uplifts to the controls & security policy
 Cracked 70% MD5 encrypted passwords using brute-force, combinator & wordlist approach in hashcat

Virtual Experience Program

United Nations Academic Impact

 Social endeavor to achieve digital wellbeing amidst deteriorating mental health due to technology abuse
 Reached out to 80+ SMEs from Chennai; Orchestrated #IAmDigitallyResponsible Challenge (reach 8k+)

United Nations and MCN Millennium Fellowship 2020

E-Cell IIT Madras

 Nominated to spearhead a 3-tier team of 15+ members & orchestrate the first virtual Youth Conclave
 Introduced 8+ new events with 3.5k+ attendees (200% increase YoY); Managed an overall budget of INR 5L
 Restructured strategies to engage 10,000+ audience through digital media & events during COVID lockdown
 Incremented freshmen participation by 120%(200+ mentees) via E-Buddy sessions & thematic counselling

Head | Entrepreneurship Club

 Handled Entrepreneurship Awards, aimed to felicitate best on-campus entrepreneur (50% growth YoY)
 Led E-Shows & managed a footfall of 2k+ participants; ideated major networking events and meetups
 Articulated 10+ yearlong activities (including E-Auction) to demystify the notion of entrepreneurship

Manager

The ePlane Company

 Converted 6 deals post revamp & modelling the prototype surveillance hexacopter to meet DGCA regulations
 Achieved 20% weight reduction, maintaining 1kg payload capacity, updating from plate to rod-based design
 Reduced cost by 15%, avoiding machining & material costs, using 3d printed parts for low stress members

Manufacturing and Design intern

Shaastra, IIT Madras

Envisage Coordinator

Mechanica IIT Madras

Coordinator

I am driven by an unwavering passion for continuous learning and expanding my expertise in the field of Machine/Deep Learning. Each day, I immerse myself in new research and advancements, making it my hobby to constantly grow and embrace the role of a perpetual student.
- Graduate student from the Indian Institute of Technology Madras, specialising in Data Science/Machine Learning.
- Completed renowned Stanford course by Andrew Ng, gaining a solid foundation in Machine Learning and setting the stage for my journey into AI.
- Deepened my knowledge and practical skills through the Deep Learning specialization by deeplearning.ai, fueling my passion for the field.
- Received Millennium Fellowship of the United Nations Academic Impact and MCN, selected from 15,000 applicants worldwide, leading 1,000+ Fellows in a technology-focused health and wellbeing initiative.
- Led the Entrepreneurship Club at IIT Madras, overseeing a 3-tier team to promote entrepreneurship and organize a pan-India summit. E-Summit, India's first ISO-certified summit, offered workshops, keynotes, competitions, and valuable resources for aspiring entrepreneurs.
- Expanded my academic pursuits by extending my bachelor's program to pursue a master's degree in Data Science, furthering my expertise.
- Worked at Microsoft India, contributing to the ranking problem for the People Also Ask (PAA) block of the Bing Search Engine, implementing cutting-edge ideas from research papers.
- Currently serving as a part-time NLP engineer at Cymorg Inc., where I've gained valuable startup experience and focused on creating ML-powered products.
- Actively engaged in the Full Stack Deep Learning course by UC Berkeley alums, honing my skills in building end-to-end ML systems.
- Conducting research with Prof. Arun Tangirala in the Question Generation space using NLP, exploring innovative applications of natural language processing.
- Proud achievement: Developed a highly successful gibberish detector for the English language, with 975k downloads in a month and reaching the 3rd page of most downloaded models worldwide on HuggingFace, competing with models from industry giants like Google, Microsoft, Meta, and others.

taplio

The Battle

Abhishek Thakur

Madhur Jindal

Followers

Posts

Engagements

Likes

Ready to grow your LinkedIn presence?