Kshitij Sharma
AI is analyzing your overall score…
Identifying your key strengths…
Evaluating your skill match against the job requirements…
Assessing your cultural and operational fit
AI is analyzing your overall score…
Identifying your key strengths…
Evaluating your skill match against the job requirements…
Assessing your cultural and operational fit
AI & ML | Building AI Agents, Predictive Systems, and Recommendation Engines | LLMs • Bayesian ML • C++/Python
I’m a Data Scientist and Machine Learning Engineer with over a decade of experience building AI systems across automotive, personalization, and enterprise domains — from embedded AI and predictive modeling to large-scale recommendation engines and LLM-powered agents. My work bridges research and production, combining advanced deep learning, LLM architectures and Bayesian inference with real-world deployment across Python and C++ environments. I’ve designed and led projects ranging from AI-driven vehicle intelligence systems and real-time prediction models to personalized recommendation engines used by tens of millions of users. I’m passionate about building AI that works in production — systems that are interpretable, scalable, and impactful. My experience spans the full ML lifecycle: prototyping, optimization, deployment, testing, and long-term maintenance, supported by strong foundations in software engineering and data infrastructure. Key Expertise & Focus Areas: - AI & Machine Learning: Generative AI, Large Language Models (LLMs), Retrieval-Augmented Generation (RAG), Fine-Tuning, Bayesian Inference, Causal Inference, Recommender Systems, Forecasting, Optimization, MLOps - Software Development: Python, C++, SQL/NoSQL, Spark, AWS, Docker, Django - Applications: Embedded AI, Voice & Chat Agents, Search Personalization, Music Recommendations, Predictive Modeling, Analytics Dashboards
North Carolina State University
Master's degree, Computer Science
January 1, 2014 – January 1, 2016
National Institute of Technology Silchar
Bachelor's degree, Computer Science and Engineering
January 1, 2007 – January 1, 2011
Dach & Nona
Founding AI Engineer | Systems Architect
July 1, 2025 – Present
Gurugram, Haryana, India · On-site
Mercedes-Benz Research & Development North America, Inc.
Lead Data Scientist (Consultant)
April 1, 2021 – June 1, 2025
Gurugram, Haryana, India · Remote
Wynk Limited
Lead Data Scientist
April 1, 2020 – March 1, 2021
Gurugram, Haryana, India · On-site
Mercedes-Benz Research & Development North America, Inc.
Machine Learning Engineer
August 1, 2016 – May 1, 2019
San Francisco Bay Area · On-site
SAS
Software Engineer, Research Intern
June 1, 2015 – April 1, 2016
Raleigh-Durham-Chapel Hill Area · On-site
Publicis Sapient
Software Engineer
November 1, 2011 – June 1, 2014
Gurugram, Haryana, India
Deltecs InfoTech
Software Engineer, Intern
May 1, 2010 – July 1, 2010
Mumbai Metropolitan Region
Dach & Nona AI Agent
April 1, 2025 – Present
Developed a domain-specific AI agent for a café using FAISS, GPT-4, and custom ML models to handle menu queries, personalized recommendations, and analytics-driven suggestions. Integrated proactive behavior, semantic caching, and feedback loops to optimize performance and user engagement. Built with Django and AWS, the system was integrated with WhatsApp and the inventory management platform to automate database queries and enable faster dashboards and analytics.
Document summarisation and multi-label classification, NLP
January 1, 2020 – January 1, 2020
Developed a model to capture important sub-sentences from a document and assigning the sub-sentences with relevant categories. Devised a Bidirectional LSTM with attention and a Convolution Neural Network architecture with GLoVe word embedding. (Tensorflow,Python)
Image and Text Labelling
April 1, 2016 – Present
This application is capable of identifying relevant articles/texts, given an input image and vice versa. Its implementation involved - Embedding images and texts into a common vector space, using the Python Tensor flow library. - Used Convolution Neural Network to extract features from images - Used Skip-gram model to extract features from texts - Extracted images and text features were used to map/transform the images and texts to a common vector space, so they could be compared to calculate similarities. - Reference: http://dl.acm.org/citation.cfm?doid=2783258.2783296
Market segmentation using Community Detection on Attributed graph
April 1, 2016 – Present
Implemented the algorithm using Newman modularity function on facebook dataset, with influence propagation 20% better than baseline K-means clustering.
NLP: Sentimental Analysis on IMDB reviews and Twitter feeds
February 1, 2016 – Present
Used Logistic regression to classify reviews on the features extracted from two models: - Bag of words model where the features were “important” words and the feature vectors were simple binary vectors. - The Doc2Vec: In this models the feature/document vectors were learned via artificial neural networks using the python gensim word2vec library. - Achieved an accuracy of 83% and 84% respectively. (Python)
Spark Streaming Twitter sentiment analysis
February 1, 2016 – Present
- Implemented a sentiment analytics application using the spark streaming API and Kafka in python. - Used Kafka queuing servicein this project to buffer the tweets before processing.
Music Recommender System using Apache Spark
January 1, 2016 – Present
A highly scalable recommender system, which was implemented using the collaborative filtering technique for implicit feedback dataset(last.fm.com), using the Alternating Least Square algorithm in Python and Spark.
Estimating Box Office Revenues using Sentiment Analysis of Tweets
April 1, 2015 – Present
We aim to develop a model which can predict the opening weekend earnings of the movies by analyzing the Revenues, Cast, Budget and Sentiment of the general masses for already existing movies. In our approach the expectation and mood of the general masses about a movie was evaluated by doing sentiment analysis on tweets of the users till a day before the release date. Sentiment of the tweets were calculated based on the weights assigned to the words in the tweets.The prediction of the movie’s first weekend collection was done using standard Multiple Linear Regression based on the sentiment factor, cast factor and the budget factor as the dependent variables.
Sentimental Analysis on Twitter feed
April 1, 2015 – Present
An application built using Node.js, Express framework to perform a sentimental analysis of real time streaming tweets, using the twitter streaming API, bootstrap.
Querying and processing trending topics on real-time stream(Tweets) using Storm/Trident
March 1, 2015 – Present
Framework built to query Heavy hitters, top-K, and frequency of topics, using the Storm/Trident for stream(Tweets) processing. Elasticsearch used for indexing and querying, Count-Min Sketch for Heavy Hitters, top-K and Bloom filter for maintaining stop words list.
Instant Messaging application using Node.js
February 1, 2015 – Present
Multi-Client chat-room with persistent chat log handling: A multi-client chat room developed using Node.js and Redis, capable of persistent in-memory data handling, populating prior chat logs in real-time.
Cultural Fit Analysis
The candidate's project portfolio is heavily skewed towards AI/ML engineering and research, with a strong focus on developing models and systems. While these skills are highly relevant to a 'Data Analyst' role, the depth of experience leans more towards a 'Data Scientist' or 'ML Engineer'. The projects demonstrate a strong inclination towards innovation and complex problem-solving, which could be a good cultural fit for a dynamic, research-oriented environment. However, the direct alignment with typical 'Data Analyst' responsibilities (e.g., dashboarding, reporting, business intelligence tools) is less explicit. The diversity of projects and roles (from embedded systems to full-stack AI platforms) indicates adaptability and a broad technical interest.
Soft Skills & Operational Fit
The candidate's project descriptions and work history suggest strong problem-solving abilities and a proactive approach to developing innovative solutions. The experience at Dach & Nona as a Founding AI Engineer indicates leadership, architectural thinking, and end-to-end project ownership. The descriptions imply an ability to work on complex, multi-faceted problems and integrate diverse technologies. However, without specific psychometric or communication test results, a direct assessment of soft skills like teamwork, stress handling, or direct communication clarity is limited.