Sudeep Nellur Portfolio

💻 Experiences 💻

Associate Product Engineer

TheMathCompany, Bangalore

Jan 2023 - Present

I contributed to an internal team, focusing on the development of management tools. My responsibilities included utilizing FastAPI, implementing PostgreSQL with SQL Alchemy, and integrating Azure BLOB for storage purposes. And also used Azure Repos for version control and documented those APIs using SwaggerUI and Postman.
Implemented an ETL pipeline to extract metadata from 25+ data warehouses that were stored in Redshift, streamlining the process of loading data into PostgreSQL for the creation of a Data Dictionary. This significantly reduced the effort required for manual creation.
Developed an ETL pipeline to extract metadata from Hive Metastore and data catalog platforms like Alation. Transferred the extracted data to PostgreSQL, contributing to the creation of training data for LLM models.
Engaged in the development of an Airflow pipeline for status updates of the DAGs and notification of their status details to their designated Slack channel.

Technical Lead

Docketrun Tech Pvt Ltd, Hubli

Feb 2022 - Dec 2022

Gathered technical requirements from clients, evaluated solutions for potential bottlenecks, technical debt, and future scalability, and designed overall projects. Subsequently, conveyed the requirements to the development team, assisting them in establishing pipelines for data processing, model training, API building, and deployment.
Assisting interns and freshers in developing Python solutions, modules, and APIs, as well as providing support with database schemas and queries. Additionally, constructing GUIs for the Annotation team to enhance efficiency and streamline tasks within startup deadlines, thereby boosting team productivity.

Python Developer and ML Engineer

Docketrun Tech Pvt Ltd, Hubli

Dec 2020 - Jan 2022

Built backend APIs using Python and Flask, interacting with SQL databases such as MySQL, PostgreSQL, and AWS RDS Oracle, as well as NoSQL databases like MongoDB. Dcokerize and deployed on AWS EC2 instance.
Built an image auto-annotation system for a desired label using AWS Rekognition . This pipeline takes images, queries AWS Rekognition for coordinates of the desired label class, parses the JSON response into XML annotation format, and saves it using Python. Later, the annotations are converted to KITTI format for model training.
Built a PoC for text recognition using Python and Tesseract . Developed the frontend using Bootstrap with Flask and successfully deployed it on an AWS EC2 instance within a day.
Built Ubuntu desktop applications as part of a Video Analytics app for drawing polygon ROIs and line crossings on images. Recorded the coordinates of the desired ROI and line crossings, feeding them into the Analytics App using Python and Tkinter.
Built a data augmentation pipeline for an Indian double-lined two-wheeler's license plate image dataset. This pipeline involves taking an empty number plate image with some tilt and masking it with a desired number PNG image to create a fake license plate dataset using Python and OpenCV.
Built a motion detection pipeline where any movement in the video or stream is detected at certain frame intervals. The pipeline was initially written in Python and later rewritten in C++.
Built a Person Re-identification PoC for NEXA client, focusing on identifying individuals across different cameras and camera views. Constructed a Siamese Network with a ResNet base model using PyTorch, enabling the comparison of batches of images to identify persons. The results are then stored in PostgreSQL.
Built an ANPR (Automatic Number Plate Recognition) PoC for a Honda service center. The system detects and recognizes the vehicle type and license plate, handling all types of license plates (single-lined and double-lined). YOLO V3 is employed for vehicle type detection, YOLO Tiny for license plate detection, and a CNN for OCR. The extracted data is then stored in MongoDB.

Full Stack Web Developer

Chromosis Tech Pvt Ltd, Hubli

Jan 2020 - Jun 2020

I built a CMS web app for the Smart Campus project, where employees can register and later be assigned to specific buildings and floors for management, including contact information.
CMS system was built using Django with Bootstrap as the frontend and MySQL as the database. The management of all these components was handled with the XAMPP server, and sessions and password hashing were implemented for authentication.

🧗🏻 Side-Projects 🧗🏻

ColorCode

This is a web app where you can upload an image, and it will provide you with all the color palettes. This way, you don't have to rack your brain for the HEX code or RGB code of your favorite color from the images you've just viewed.

The app features a Clustering ML algorithm K-Means that creates clusters for similar colors. By utilizing the centroid of each cluster, we can retrieve the HEX and RGB codes. Subsequently, the app is exposed as APIs, allowing seamless integration with the frontend. Flask serves as the RestAPI framework for this purpose. Visualization is managed through Bootstrap, providing users with a smooth experience. To ensure production-grade readiness, a WSGI server using Gunicorn has been incorporated. As a final step, the app has been containerized by creating a Docker Image and is now live on the Heroku cloud server.

Demo and Documentation on

Paragraph Summarizer

This is an NLP project for text summarization, built with Flask and deployed on Heroku. It utilizes NLTK for summarizing text. The app processes extensive paragraphs and extracts only the repeating sentences deemed important for the user.

The app takes a text paragraph input from the user and iterates through each word to generate a word count frequency table. Words with higher frequencies are deemed important. Subsequently, the app analyzes each sentence, checking for occurrences of these high-frequency words. The more instances of high-frequency words in a sentence, the more important it is considered. The app is built using Python NLTK for text processing, Flask for REST APIs, and Bootstrap for the frontend. It has been deployed on Heroku for demonstration purposes.

Documentation on

Face recognition with time snapshots

This is a face recognition app that records the timestamps of the detected faces in an Excel sheet for reference. The custom model is trained using VGG16 as a base model and analytics are later incorporated using Python.

The app will utilize a system camera as a video stream to capture a feed. Each frame is then sent for prediction, where the face is cropped using Face HaarCascade. The cropped face image array is forwarded to the model for classification. The resulting prediction is recorded in the Excel sheet through Python. The data is manually collected, trained in Google Colab, and inference is performed on the local system.

Demo and Documentation on

Right Candidate

This is a dynamic Django CRUD app with MVT architecture that consists of two pages (apps). One is for applying for jobs, and the other is for displaying jobs with the ability to filter applicants based on their skillsets.

The app's application page is the candidate's application page, where the candidate fills in basic inputs such as first name, last name, skills, etc. Another page, called "Submissions," is visible for admins. On this page, admins can view all entries with respective color-coding (red, yellow, and green) indicating the suitability of candidates. The entered input is saved in SQLite, and REST APIs are served via Django, which is later accessed by the frontend using Bootstrap.

Application Page:

Submissions Page:

Documentation on

Hey Visiter 👋🏻

I am Sudeep Nellur..!!