Open to Senior Roles

Sujit
Nalawade

Senior Machine Learning Engineer specializing in Computer Vision, Deep Learning, and scalable AI inference systems — building things that see and understand the world.

7+
Years Exp.
1000+
Cameras Live
200+
Retail Registers
✉ Get in Touch ↗ LinkedIn
Sujit Nalawade

Sujit Nalawade

Senior ML Engineer @ Adagrad AI
📍 Pune, India
🎓 PGD Data Science — IIIT Bangalore
🎤 NVIDIA GTC Speaker (2021, 2022)
📧 technosujit25@gmail.com
📱 +91 8329132283
// About Me
Building AI that
sees the world

I'm a Senior ML Engineer with 7+ years of experience building and deploying machine learning systems at scale. My work sits at the intersection of computer vision, deep learning, and production engineering.

From real-time video pipelines processing thousands of camera feeds to inference servers handling high-throughput batched requests — I specialize in taking research-grade models and turning them into robust, production systems.

I've spoken at NVIDIA GTC twice, sharing work on object detection at 1770 FPS and geometric deep learning with TensorRT.

// Work History
Experience
Senior Machine Learning Engineer
Adagrad AI Pvt. Ltd. — Pune, India
Apr 2020 – Present
  • Responsible for the design, development, and deployment of Computer Vision and NLP pipelines for solving real-world use cases.
  • Close collaboration with the team to brainstorm, design, and solve problems in a scalable fashion for large scale deployment.
Machine Learning Engineer (NLP) — Freelance
Lineslip Solutions Pvt. Ltd. — New York City
Dec 2018 – Apr 2020
  • Trained word embedding generation algorithms using Word2Vec and Gensim for words from insurance documents.
  • Trained Bidirectional LSTM and BERT for named entity recognition in docs, improving F1 score from 0.60 to 0.85.
  • Trained a Doc2Vec-based document classifier with an F1 score of 0.93 for large document classification.
Software Engineer
Xoriant Solutions Pvt. Ltd. — Pune, India
Jun 2016 – Mar 2020
  • Developed a data masking library for masking out personal information fields from large XML documents using C/.NET.
  • Designed a web interface for payload management using Angular and Node.js.
  • Certified in Microsoft Azure — "Developing Microsoft Azure Solutions".
// Selected Work
Projects
🎥

DNSTAR

High-performance deep learning inference pipeline for real-time video processing, supporting object detection (YOLOX, YOLOv8) and classification models. Custom Python/C++ bindings for analytics. Deployed across global sites spanning 1000+ cameras on dGPU.

👤

REID & MATCHING

Cross-camera tracking microservice matching pedestrians via ReID feature vectors and cosine similarity. Trained ViT-S16/B16 with triplet loss on a curated 5K-identity dataset, reaching 0.9 mAP and 0.93 Rank-1. Cut runtime from 72 hrs to 3 hrs via multiprocessing.

🛒

CHECKOUT ACCURACY

Video pipeline using a custom YOLOv5-OBB model to detect unbilled goods in shopping carts from overhead cams at checkout. Deployed across 200+ registers at one of the largest retail chains in the US and Canada.

🚗

GATEGUARD

Real-time vehicle number plate identification pipeline (YOLOX, YOLOv8) with custom event aggregation. Deployed across 100+ sites in 10+ Indian cities, including tolls, malls, and housing societies.

🏠

NOBI ENGINE

Oriented bounding box detection pipeline (YOLOv5-OBB, 0.9 mAP@50) identifying elderly resident actions in care rooms, with custom OBB support added to DeepStream's Nvinfer component. Deployed on Jetson Nano, TX2NX, and NX edge devices.

⚡

TRITON INFER ENGINE

Inference API server fronting NVIDIA Triton with batched multi-model inference support, serving YOLOX, YOLOv5, YOLOv5-OBB, YOLOv8, and ResNet-18 as the interface between end users and the inference server.

💬

TAXBOT

RASA-based chatbot answering Indian tax-filing queries for a Mumbai-based financial company, combining a custom-trained conversational bot with similarity search for out-of-domain Q&A. Deployed on AWS EC2.

🏪

POSEIDN

Video pipeline using YOLOv8 pose estimation to measure customer interaction in retail store sections, helping optimize layout and shopping experience. Deployed across multiple retail stores in the EU.

📦

SOPHIA

Inference pipeline combining a custom YOLOv8 detector with Microsoft Azure Vision API to extract and match company-specific codes on containers and trailers, deployed on managed Azure Triton and integrated with other microservices.

🏭

WAREHOUSE ANALYTICS

Analytics product tracking inflow/outflow of cement bags in a warehouse using a YOLOv8-backbone detection model trained with Azure Vision APIs, deployed for cloud inference via FastAPI.

// Technical Stack
Skills

Expertise

Computer VisionObject Detection & TrackingImage ClassificationNLP

Languages

PythonC++C

ML / Vision

PyTorchPyTorch LightningCUDATensorRTONNXRuntimeScikit-LearnOpenCV

NVIDIA Stack

DeepStreamTriton Inference ServerGStreamer

NLP / LLM

HuggingFace TransformersSentence TransformersRASA

Backend / DevOps

Django RESTFlaskStreamlitDockerDocker ComposeGit / GitHub ActionsPostgreSQLMySQL
// Background
Education
Post Graduate Diploma in Data Science
International Institute of Information Technology, Bangalore
Sep 2018 – Sep 2019 · Score: 3.16/4
Bachelor of Computer Engineering
University of Pune, Pune, India
Jun 2012 – Jul 2016 · First Class with Distinction
// Public Speaking
Conference Talks
GTC 2022
Save Thousands of Man-hours with Geometric Deep Learning and TensorRT
NVIDIA GPU Technology Conference
GTC 2021
Object Detection at 1770 FPS: Designing CNNs that Scale Using DeepStream SDK
NVIDIA GPU Technology Conference