Nemish KanwarSenior Data Scientist
Profile

Passionate Data Scientist & ML Engineer with ~10 years of experience in building and deploying large-scale machine learning systems. Specialized in architecting data-intensive models, solving scalability and performance challenges, and implementing MLOps best practices for seamless model deployment and monitoring. Expertise in Natural Language Processing (NLP), Deep Learning, and engineering high-performance data pipelines. Experienced in designing robust, production-grade ML infrastructure in both startups and enterprise environments. Strong leadership in team setup, recruitment, and mentoring, with a track record of optimizing ML workflows from research to production.

Education

Birla Institute of Technology and Science, Pilani

BE (Hons) in Mechanical Engineering
2012 – 2016
Skills
Languages: Python, R
Framework: Tensorflow, Pytorch, Keras
Large Language Models: Llama 3, Mixtral, Gemma, GPT-4
Embeddings & Vector Search: Word2Vec, GloVe, FastText, FAISS, ChromaDB
Prompt Engineering: Few-shot learning, Chain-of-Thought (CoT), Retrieval-Augmented Generation (RAG)
Visualization: Bokeh, Plotly
Full Stack: HTML, CSS, JavaScript
MLOps: AWS SageMaker, Vertex AI, AWS EMR, Airflow, LangChain, Hugging Face Transformers
Professional Experience

Target

Senior Data Scientist
03/2022 – present | Bengaluru, India

Terraform-Driven Automation for Shepherd Platform

  • •Led end-to-end Infrastructure-as-Code (IaC) implementation for Shepherd, Target’s internal workflow-creation platform, replacing UI-based workflow setup with Terraform-driven deployments
  • •Built Terraform modules and standardized code patterns to enable consistent multi-environment (dev/stage/prod) deployments across multiple Shepherd teams.
  • •Designed and implemented a user-level dev deployment workflow, enabling rapid iteration, testing, and self-service deployments for developers
  • •Collaborated with cross-functional teams to consolidate deployment practices and streamline Shepherd adoption through code
  • •Onboarded and trained teams on Shepherd IaC, accelerating migration from UI-driven configurations to version-controlled, scalable infrastructure

    • Co-occurrence & BiCM-Based Modeling for Line-Item Grouping

  • •Engineered a large-scale co-occurrence model, incorporating concepts from Bipartite Configuration Models (BiCM) to study complement and substitute effects across 8M+ transactions and 200K+ SKUs, supporting exploratory pricing research
  • •Adapted BiCM-inspired mathematical formulations—originally designed for single-machine Python—into a distributed, production-grade PySpark implementation, enabling efficient computation on large retail datasets
  • •Designed and implemented a scalable similarity metric to capture complementary and supplementary relationships while normalizing for large basket sizes to ensure meaningful associations
  • •Optimized the full Spark workflow through strategic partitioning, caching, and broadcast joins, significantly improving performance and reducing job runtimes.
  • •Built a resilient data processing pipeline capable of operating directly on raw, unfiltered two-month transaction windows, demonstrating algorithmic robustness and scalability
  • •Delivered foundational insights for line-item grouping within pricing workflows, enabling deeper exploration of item relationships for future ML and pricing initiatives

    • Project: Offer Forecasting and Optimization

  • •Offer Types Analysis: Analyzed BOGO, percentage discounts, and mandated multiples to optimize SKU discount depths.
  • •Elasticity Modeling: Developed mixed effects models for class-level elasticity to predict optimal discount depths.
  • •Optimization: Applied elasticity calculations to recommend discount depths and offer types, boosting promotion effectiveness and ROI.

    • GenAI Hackathon Project

  • •Engineered a GenAI solution translating user-provided natural language queries into HQL scripts, executed across 1200 tables with approximately 50,000 unique columns.
  • •Developed NLP prompts for Mixtral-7b, a large language model (LLM), employing the Jina sentence embedding model.
  • •Enhanced LLM prompt context by leveraging embeddings, resulting in improved model performance and response accuracy
    • Courses

    Machine Learning DevOps Engineer

    Udacity

    Deep Learning Specialization

    Coursera

    Deploying Machine Learning & NLP Models with Docker

    Udemy

    Full Stack Engineer (ongoing)

    Codeacademy
    Awards

    Second prize at Target Hackathon

    2024

    First Prize at Draup Hackathon

    2020

    Draup Fast Tracker Award

    2018

    Draup Fast Tracker Award

    2017
    Social Projects

    "Empowering Voices" - Building a Central Platform for Bringing Together Success Stories of Udaayan Care NGO's Girls

    2021
  • •Led the creation of a book by capturing and compiling success stories of Udaayan Care NGO's girls
  • •Developed a comprehensive data framework, orchestrating end-to-end processes for a compilation of 60 impactful stories, utilizing open-source and free tools for development

    • Student Progress dashboard for NIRMAAN NGO

    Prepared a dashboard on Google Data Studio and Google Forms for ground-level monitoring of student education health

    Interests
    Long distance cycling|Singing, open mics|Gardening/composting

    goSwift Logistics

    Senior Data Scientist
    07/2021 – 03/2022 | Bengaluru, India

    Managed Data Science team and led development of data models and pipelines using Airbyte and BigQuery. Designed, developed, and deployed ML models as microservices on Google Cloud Run.

    Draup

    Senior Data Scientist
    08/2017 – 07/2021 | Bengaluru, India

    Classifying News into Signals

    ML Models: Trained classifiers like Random Forest, SVM, GRU, LSTM, CNN, and CNN-GRU using TensorFlow, PyTorch, and embeddings (Word2Vec, GloVe). Explored Universal Sentence Encoding for multinomial classification with unbalanced data.

    Model Training: Managed weekly training on AWS SageMaker, deploying models to Databricks via Docker, Lambda, and CloudWatch automation.

    Signal Validation: Developed a robust validation process, including team reviews and signal retagging.

    Executive Matching Models

    Engineered a profile matching system for LinkedIn against Crunchbase, Relsci, and Twitter using Jaro-Winkler, Doc2Vec, and distance metrics.

    Employed Random Forest classifier for cross-platform similarity search.

    Implemented a categorization strategy based on confidence levels for efficient match prioritization.

    Capgemini

    Associate Consultant - Data Science and Analytics
    07/2016 – 06/2017 | Bengaluru, India

    Worked on document classification/​processing for a European bank using OCR, ML, and NLP techniques.

    Personal Projects

    Integrated ChatGPT to conversate as natural language with own data in backend

    2023

    Developed a versatile chat bot incorporating GPT-3.5, providing real-time insights on queries based on own data

  • •Integrated ChatGPT for pattern-based question matching and fixed-answer template generation.
  • •Employed fuzzywuzzy to ensure accurate query cases and word matches for input dataframe values.
  • •Utilized ChatGPT API to transform natural language inputs into SQL queries seamlessly using prompt engineering.
  • •Implemented pandasql to efficiently query narrowed-down datasets based on generated SQL.
  • •Orchestrated the integration of a shortened dataframe and natural language inputs to generate coherent responses.

    • Job description extraction from a custom company JD page

    2022

    Extracting relevant content from a custom company career page

  • •Dividing a HTML webpage to relevant blocks using DOM structure
  • •Classifying a block into relevant and non-relevant block
  • •Achieved 0.93 AUC on the random forest model