Hi 👋, I'm Susheel R N!
I am a passionate Senior Data Scientist specializing in NLP, Generative AI, and Computer Vision, with a strong foundation in Machine Learning and Deep Learning.
With 6+ years of experience, I have worked across the AI lifecycle - from core model development and system architecture to optimized deployment on edge devices and cloud platforms. My experience spans both software and hardware-aware AI engineering, with a focus on building solutions that balance accuracy, latency, scalability, and resource efficiency.
My core expertise includes LLMs, real-time video analytics, AI-driven automation, model optimization, GPU acceleration, and high-performance inference. I enjoy tackling complex problems where research, engineering, and deployment come together to create practical business impact.
Driven by an entrepreneurial mindset, I continuously explore emerging technologies and transform promising AI concepts into reliable, scalable applications for real-world use cases.
Technical toolkit
Tools and platforms I use to build, optimize, and deploy AI systems.

Python

TensorFlow

PyTorch

Keras

ONNX

Scikit-Learn

SciPy

NumPy

Pandas

OpenCV

YOLO

Ultralytics

Hugging Face

OpenAI

LangChain

LangGraph

Apache Airflow

Apache Kafka

Elasticsearch

Logstash

Kibana

PostgreSQL

MySQL

SQLite

AWS

Google Colab

GitHub

Bitbucket

Jupyter

PyCharm

Spyder

Pytest

JavaScript

Scrapy

Selenium

SpaCy

FastAPI

Arduino

Raspberry Pi

OpenGL

Unity

CrewAI

Python

TensorFlow

PyTorch

Keras

ONNX

Scikit-Learn

SciPy

NumPy

Pandas

OpenCV

YOLO

Ultralytics

Hugging Face

OpenAI

LangChain

LangGraph

Apache Airflow

Apache Kafka

Elasticsearch

Logstash

Kibana

PostgreSQL

MySQL

SQLite

AWS

Google Colab

GitHub

Bitbucket

Jupyter

PyCharm

Spyder

Pytest

JavaScript

Scrapy

Selenium

SpaCy

FastAPI

Arduino

Raspberry Pi

OpenGL

Unity

CrewAI
Live updates
Latest posts and articles
Medium
Positional Encoding, Illustrated: A Step-by-Step Mathematical Walkthrough
Medium
Part 2- How Tokenization Works in LLMs: BPE, WordPiece, and SentencePiece Explained
Medium
Part 1- How Tokenization Works in LLMs: From Raw Text to Tokens
Latest LinkedIn posts
View my latest professional updates and recent activity on LinkedIn.
Medium
Positional Encoding, Illustrated: A Step-by-Step Mathematical Walkthrough
Medium
Part 2- How Tokenization Works in LLMs: BPE, WordPiece, and SentencePiece Explained
Medium
Part 1- How Tokenization Works in LLMs: From Raw Text to Tokens
Latest LinkedIn posts
View my latest professional updates and recent activity on LinkedIn.
Experiences
Jan 2026 - Present
Senior Data Scientist at Tredence Inc
Future Forecasting: Building demand forecasting solutions to help businesses anticipate future requirements and make data-driven decisions.
CPG Domain: Developing classical machine learning models to predict competitor behavior, own-price and cross-price elasticity, and other commercial drivers for pricing and growth analysis.
Jan 2024 - Jan 2026
Data Scientist at LTIMindtree
Lead Prediction: Developed an automated system to identify potential leads from user-configured websites using web scraping and LLMs, automating weekly alerts via Airflow, resulting in faster lead identification.
Risk Prediction: Contributed on a model using Advanced RAG and LLMs agents to analyze historical data and detect risks from tenders, improving decision-making.
RFQ Automation: Built a system that automates procurement identification and assists in RFQ generation by extracting domain-specific details from 15+ files per RFQ, streamlining the process with LLM and agent-based techniques.
Search Engine: Built an LLM-powered chatbot with agent-based retrieval for 15,000+ unstructured documents, utilizing Databricks for section detection and intelligent chunking to improve search accuracy and efficiency.
These systems collectively reduced tender processing time from 30 days to just 1 day, accelerating decision-making and optimizing capital utilization in Tenders.
Mar 2023 - Dec 2023
ML Engineer at Responsive
Document Analysis: Built a system leveraging open-source NLP libraries to extract key information from documents, including dates, sections, locations, organizations, job titles, laws, monetary values, names, percentages, contact details, URLs, and domain-specific keywords.
Document Summarization: Developed a hybrid summarization approach combining BERT-based extractive techniques with GPT models to generate concise, high-quality summaries that effectively capture critical content.
Auto Completion: Developed an AI-driven system using GPT that dynamically generates responses based on user history and context, enhancing efficiency.
Nov 2020 - Feb 2023
Associate ML Engineer at Responsive
Response Assistant using RAG: Developed GPT-powered features for first-draft generation, summarization, proofreading, and readability enhancement using RAG, enabling users to speed up RFQ response creation.
Enhanced Recommendation Engine using RAG: Upgraded the retrieval pipeline by integrating RAG with Ada embeddings, leveraging a larger embedding size than BERT. This improved accuracy and relevance for RFP suggestions by 60%.
Answer Snippet Extraction: Built a SQuAD fine-tuned BERT model to retrieve the most contextually relevant portion of an answer from large unstructured content, eliminating the need for manual document navigation.
Important Keyword Extraction: Developed an algorithm leveraging SpaCy, NLTK, TextRank, RAKE, YAKE, and KeyBERT to extract meaningful, domain-specific keywords, enhancing search functionality and information retrieval.
Jun 2020 - Oct 2020
Student Intern at Responsive
Recommendation Engine: Designed an ML-powered retrieval pipeline using BERT vector embeddings and similarity search to suggest the most relevant questions from a database, improving efficiency in responding to RFPs.
Requirement Detection from Documents: Fine-tuned BERT and T5 for sentence classification, automating question and statement identification, and achieving 85% efficiency improvement by reducing manual document processing.