Vaishak Makam

An enthusiastic and diversely skilled Artificial Intelligence Application and Research Engineer with 7+ years of experience in Deep Learning, Machine Learning, Image Processing, Recommendation engine, Programming. Experienced in building end-to-end ML/DL pipelines for various POCs. Worked with a wide variety of clients and their datasets related to automotive, home appliance, kitchen appliance, health care, entertainment, and many more.

• Young Leads Program: Identify potential leaders and build the capabilities of young talent at Tata Elxsi (~ 7 months)
• Customer Appreciation Award: The client expressed high satisfaction with the completed work and achieved results.
• Bravo Award: Both the team lead and manager were extremely pleased with the work done.
• Neural networks and deep learning by Andrew Ng, deeplearning.ai, Coursera
• A Crash Course in Data Science by Johns Hopkins University, Coursera

Road Safety Management Systems:

• Developing a road safety management system tailored for a prominent Indian manufacturing industry, aiding traffic marshals in detecting individuals breaching traffic regulations. The system oversees various infractions, including cellphone usage, failure to wear seat belts, unauthorized overtaking, speeding, and jaywalking.

Smart Defect/Damage Analysis:

• Developed a defect/damage detection module for a leading German Automotive Manufacturer aimed at identifying gear damages in vehicles. This involves comprehending various types of damages or defects that may arise due to stress, strain, friction, and related factors. Implemented object detection techniques to identify such issues.

Recommendation Engine:

• Creating a recommendation engine for a new OTT platform of a Mumbai based theatre to suggest content to users by analyzing their viewing history and demographic data. This engine categorizes users into "cold start" and "normal" users, tailoring recommendations accordingly. For normal users, personalized rails like "Because you watched" and "Recommended for you" are provided, while generic rails such as "Trending now" and "Best of artists" are suggested to all users.
• The system retrieves user watch details at predefined intervals from Kafka, processes this data as input for the recommendation engine, and stores the output in Elasticsearch across multiple indices.

Face Similarity Module:

• Developed a face similarity algorithm for a prominent US-based Technology Giant to analyze and evaluate the resemblance between two provided faces. This involves extracting facial embeddings using "FaceNet" technology and comparing these features to generate a similarity score utilizing various distance metrics.

Proctoring Module (to oversee students during online exams):

• Developing a face recognition module utilizing FaceNet to authenticate student identities. Designing a mobile phone detection module with a low-footprint algorithm. Combining these modules into a cohesive software pipeline.

ML Based Odor Classification:

• Developed a solution for a prominent Japanese Home Appliance Major to classify odors as Pleasant or Unpleasant. This involved investigating and understanding data from piezoelectric sensors, Conducting Exploratory Data Analysis (EDA) to visualize data for enhanced understanding, Selecting appropriate Machine Learning (ML) algorithms for classification, Fine-tuning algorithm parameters using GridSearch to attain desired results, Implementing Principal Component Analysis (PCA) for dimensionality reduction.

Regression Analysis (Analysis of strain on automobile wheels):

• Led the comprehensive feasibility study to determine the viability of identifying wheel strain using strain gauge sensor data supplied by the customer, a prominent German Automotive Manufacturer. Examined feature importance and correlation to assess the impact of strain gauge on wheel bearings. Quickly prototyped using diverse regression algorithms to validate the hypothesis. Engaged with Subject Matter Experts (SMEs) on the client's side to gain deeper insights into the data. Produced a detailed report outlining the findings and conclusions, earning high praise from the customer.

Smart Home Appliance Products by Japanese Home Appliance Major:

• AI aided robo vacuum cleaner: Detecting and identifying various small household objects in real-time to enable the robo-vacuum cleaner to avoid obstacles which were missed by the IR sensors. This involves implementing an object detection model, such as YOLO, specifically trained for recognizing small to tiny objects.
• Smart Laundry Systems: Developing a system to detect and identify the type of clothing and its owner, with results displayed on a graphical user interface (GUI). The laundry system is divided into several subsystems or modules, including "Clothes Detection", "Clothes Recognition", and "Owner Identification." Training and testing involved object detection techniques like SSD MobileNet for clothes detection, classification methods for clothes recognition, and autoencoders for owner identification.
• Smart Microwave Oven: To identify various types of food recipes or ready-to-eat food packets inside a microwave oven for cooking or reheating respectively, the oven must determine the contents, volume of the recipe, or the number of food packets and adjust the time and temperature accordingly. This involved implementing a classification model for recipe classification, an object detection model for food packet detection, and utilizing image processing techniques for volume estimation.
• Health Care – Beauty and Health: Responsible for implementing FaceAPI to detect and analyze faces, generating concise dermatological descriptions to display on a mobile application. This includes face detection and facial landmark generation functionalities. Additionally, identifying areas of discolored patches and wrinkles on the face for further analysis.