Nandeshwar Gupta

Experience

Where I've built things

1. Zupee

   Gurugram, India

   1. Senior Manager, Machine Learning Engineering

      Apr 2026 — Present
      • Lead a 5–7 engineer ML team, owning roadmap, hiring, delivery and stakeholder alignment across generative AI, recommendations and platform engineering.
      • Built an AI companion agentic chatbot serving 100K DAU — natural, hyper-personalised conversations that remember past context and stay guardrailed.
      • Architected an 8-node LangGraph agent with a 6-way parallel fan-out and a bounded self-correction loop, cutting per-turn inference latency ~35% and lifting session length 14% and D7 retention by 60 BPS.
      • Fine-tuned Qwen 3.6 35B-a3b on curated production chat traffic, cutting inference cost ~65% while holding TTFT within 1s p95.
      • Built an LLM conversational memory system (semantic, episodic, factual, procedural) over MongoDB + Qdrant, driving 4% higher daily active conversations and 6% higher conversions.
      • Designed a multi-layer LLM guardrails pipeline — a 24-category classifier plus an LLM-as-a-judge critic — at 70% precision / 99% recall, cutting trust-and-safety incidents by 95%.

   2. Manager, Machine Learning Engineering

      Apr 2025 — Mar 2026
      • Launched a scalable agentic LLM chatbot for astrology with RAG, SLMs and queue-based autoscaling — 100% MoM user growth, a 4-star Play Store rating and 500K+ downloads.
      • Architected an RL-based recommendation engine, evolving from multi-armed bandits to Deep Q-Networks, driving +3.5% of organisation revenue and +10 BPS retention on a 2M+ DAU platform.
      • Designed user embeddings from behavioural and demographic clustering, boosting recommendation accuracy across every downstream ML project.
      • Built SQL Smith, a graph-based natural-language-to-SQL agent that opened up self-service analytics for product and growth teams.

   3. Lead Machine Learning Engineer

      Jan 2024 — Mar 2025
      • Built and deployed real-time tournament recommendations on Apache Flink + Aerospike, cutting inference from hours to seconds and driving +4.15% revenue.
      • Improved recommendation accuracy@top4 by 6% and redesigned the architecture to resolve cold-start; parallelised retraining with Airflow, taking training cycles from hours to minutes.
      • Shipped a multi-armed bandit pricing optimisation system, delivering +8% revenue uplift.

   4. Senior Machine Learning Engineer

      Sept 2022 — Dec 2023
      • Spearheaded the end-to-end ML lifecycle framework and real-time prediction stack — training, tracking, deployment, monitoring — from scratch.
      • Delivered the XGBoost tournament recommender, boosting revenue by 4% and accuracy@4 by 10%.
      • Led generative-AI proofs of concept: LLM quiz generation, Indian-dialect TTS for game commentary, marketing video generation and SageMaker GPU training optimisation.

2. V-Mart Retail

   Gurugram, India

   1. Senior Manager, Data Science

      Feb 2021 — Sept 2022
      • Built a matrix-factorisation recommendation engine, lifting platform revenue by 11%.
      • Developed churn prediction (LightGBM) and customer segmentation models, improving retention and personalisation.
      • Created ARIMA forecasting for the supply chain, materially improving demand planning through COVID-19 volatility.
      • Designed scalable ML pipelines on AWS using Airflow and MLflow for reproducible training and deployment.

3. Vasitum

   Gurugram, India

   1. Manager, Data Science

      Sept 2020 — Feb 2021
      • Built recommendation engines and an NLP chatbot (RASA, NER, intent classification) to lift engagement and conversion.
      • Optimised job search with Elasticsearch and shipped NLP/DNN models for spam-job classification.
      • Delivered end-to-end ML pipelines, including database modelling and migration to serverless AWS (Lambda, Fargate).

4. Quick Company

   Gurugram, India

   1. Data Scientist

      May 2018 — Aug 2020
      • Built large-scale NLP and computer-vision models: NER (spaCy), legal text classification (Word2Vec), captcha recognition (CNN, OpenCV) and image classification (PHOG).
      • Developed an automated, event-driven scraping framework and used Neo4j for graph-based data linking across legal entities.

Toolkit

What I work with

Generative AI & LLM

RAG Agentic Frameworks LangChain LangGraph LlamaIndex Prompt Engineering Fine-tuning (LoRA/PEFT) SLMs Vector DBs (FAISS, Pinecone, Qdrant) Tool-calling Guardrails LLM Observability

Machine Learning

Recommendation Systems Reinforcement Learning (DQN, MAB, RL-Lib) Deep Learning NLP (NER, Intent, Classification) Computer Vision (CNN, OpenCV) Forecasting (ARIMA) Matrix Factorization LightGBM Clustering & Embeddings

ML Engineering / MLOps

MLflow Kubeflow Airflow SageMaker Feature Stores Model Registry A/B Testing CI/CD for ML Real-time Inference (Flink, Aerospike) Monitoring & Drift Detection

Cloud & Infra

AWS SageMaker Lambda Fargate EKS S3 Athena Docker Kubernetes Jenkins Bitbucket Pipelines

Languages & Frameworks

Python SQL Bash JavaScript PyTorch TensorFlow Scikit-learn Hugging Face FastAPI Flask

Leadership

Team Management (5–7 engineers) Hiring Roadmapping Cross-functional Partnership (Product, CRM, Marketing) Mentorship

Education

Where I studied

• B.Tech, Computer Science Engineering

  Manav Rachna International University, Faridabad

  2014 — 2018

Off the clock

What drives me