Hello! Myself Piyush Jain. I am a Deeplearning, Computer Vision and Robotics Engineer.
I love the idea of exploring how a MACHINE is LEARNING and EVOLVING and how I can use my adaptive problem-solving skills to help these ARTIFICAL brains multi-fold their INTELLIGENCE. These ARTIFICIAL brains never cease to amaze me, especially with the advent of ARTIFICIAL INTELLIGENCE, my fascination for these machines has grown manifold. So if you've opportunities for me, or have crazy product ideas to discuss, always feel free to reach me via email. Check out my resume. I'd be very happy to connect.
Oct 2021 - Present
• Developed a RAG - Mixtral-like bot leveraging OpenAI's GPT-4 technology, providing specialized parking assistance with enhanced user interaction and efficiency.
• Enhanced multi people detection and tracking algorithms, optimizing model deployment across 40+ cameras. Smart grocery stores akin to Amazon go.
• Managed end-to-end model development lifecycle from data preparation to deployment. Deployed to 4 locations spread across Finland and Sweden.
May 2021 - Sep 2021
• Expanded canonicalization algorithm by 10x handling over 15 million data points and reduced the search time by 5x using concept mapper.
• Architected, maintanied and remodeled a whole end to end project in PLAY framework servicing over 100+ customers.
• Engineered a Auto correct algorithm using Seq-Seq LSTM model and currently installed in product servicing 200+ clients.
Oct 2020 - March 2021
• Implemented visual perception and deployed to a robot for it to traverse through buildings and find fire exits.
• Direction and angle of arrows were detected using Retina-Net with an accuracy of 87.6%.
June 2020 - Nov 2020
• Provided analytics on sports like Boxing, Diving, and Football for US Olympics using computer vision.
• Improved the accuracy of pose estimation by 4.22% for combat sports analysis by using Detectron2 and Denspose.
• Implemented various image processing algorithms on boxing to obtain stance classification, distance travelled, player
detection, with an accuracy of 93.4%.
July 2020 - Aug 2020
• Developed 12 chapter’s of tutorials and content using cognimates and implemented programs in the field of CV and NLP.
• Course has been taken by 250+ students so far with 4.65 average rating.
May 2019 - Feb 2020
• Developed a full sized Smart ICU bed with python and HTML running on raspberry pi and was deployed in a hospital.
•Designed Smart blub using EasyEDA and Automatic Plant watering system using C. Performed testing, documentation,
verification, productization and maintenance of algorithms along with s/w modules that use them.
June 2019 - July 2019
-Development of smart energy meters. -Research of quantitative stratergies in production of a product.
2021 - 2023
Technische Universit ̈at Berlin and Aalto Univeristy
Courses : Machine Learning, Natural Language Processing, Computer Vision, Robotics, Parallel Programming, Deep learning
2017 - 2021
Sri Venkateshwara College of Engineering
College bolstered me to build better knowledge in areas of Robotics and Embedded Systems. Gave me better skills on writing, communicating, studying, and critical thinking. Courses : Digital Image Processing, Digital Signal Processing, Data Structures, Robotics, Statistics
Developed a system to assist chartered accountants with summaries, query responses, and document retrieval. Enhanced efficiency and accuracy in handling complex accounting queries and document retrieval by leveraging embedding-based search.
- Tools Used : Langchain, OpenAI, Streamlit, SQL
The main task was to build a model that predicts the human activities such as Walking, Sitting, Standing or Laying on the basis of data collected from sensors (Accelerometer and Gyroscope) in a smartphone. ‘3-axial linear acceleration' from accelerometer & '3-axial angular velocity' using Flutter and stored it on Firebase. Performed various algorithms like Random Forest, Convolution Networks and Convlstm2d to measure the accuracy .
- Skills Used : Keras, Sklearn, Numpy, Matplotlib, Python
The main task was to build a Hydroponic model that could grow plants. LSTM was used to predict the outcome of the plant, LeNet5 was used to detect if the fruit is rotten or not. NodeMCU and Adafruit were used to automate the whole system.
- Skills Used : Pattern Recognition, LSTM, Python, IOT
The main task was to build a model that generates new keyboard music notes based on training data music. Notes and Chords were converted to word vectors which were fed into a LSTM network.
- Skills Used : Keras, Pandas, MIDI, Python
The main task was to devise the best algorithm to predict user ratings for films. I designed a Restricted Boltzman Machine from scratch using python and predict if a person likes the movie or not.
- Skills Used : Pytorch, Numpy, Mathplotlib, Pandas, Python
The main task was to build a program that can convert contents present in an image to excel sheet(csv format). Used OpenCV's wide range of tools like contors to find the table and performed OCR using tesseract API. To know the whole story behind my motivation to build this program Click Here.
- Skills Used : OpenCV, Numpy, CSV, Python, API
The main task was to detect whether a person is wearing a mask of not and if people are following social distancing. Transfer learning was performed on MobilenetV2 using 1400 images(equally distributed). For infernce, face was detected using a caffe resnet model and trained model was applied to face ROI.
- Skills Used : Transfer Learning, Caffe, OpenCV, Python
The main task was to build a model that completely generates new car designs for Automobile industry. The data was fed into GAN (generator and descriminator network) which yielded the car design. The same model can applied to design a new Dress.
- Skills Used : Keras, Skimage, Pandas,Python
Developed various projects like "People Counter APP", "Smart Queuing System", "Computer Pointer Controller" on devices like FPGA, VPU, GPU and CPU using OpenVINO toolkit.
- Skills Used : OpenCV, MQTT, Python
FMCG(Fast-Moving Consumer Goods) brands require insights into retail shelves to help them improve their sales. One such insight comes from determining how many products of their brands’ are present versus how many products of competing brands are present on a retail store shelf. Transfer learning was performed on SSD MobileNet V2 to detect and localize the objects
- Skills Used : Tensorflow, OpenCV, Numpy, Pandas, Python
The main task was to build a All-Terrain bot which can be deployd in military having VR(360) view capabilities. Accelerometer and Gyroscope sensor values that map our head movements were sent to the RPi and Pan/Tilt was used to move the camera. It had features like weapons detection, people counter and self destruction
- Skills Used : Robotics, VR, IoT, Python, C
The main task was to set up a whole system which can be helpful to Visually Impaired. The camera inputs about the various things around a person and tells him using a earphone. YOLOV2 was used to implement the object detection and localization.
- Skills Used : Computer Vision, Embedded Systems, Python, IOT
The main task was to automate my room lights and fan. Raspberry pi and NCS2 was used to detect my face when I enter the room and turn on appliances. Google Assistant was integrated with Adafruit to perform communication around the world
- Skills Used : Embedded Systems, IoT, OpenCV, C
The main task was to help physically impared to move around. The chair has automation capabilities which can be used to turn on lights and fans.
- Skills Used : Embedded Systems, C, IoT, Python
May 2024
Coursera
-Prompting and prompt engineering. -Generative AI project lifecycle. -Pre-training large language models. -Multi-task instruction fine-tuning. -PEFT techniques: LoRA and Soft prompts
April 2020
Coursera
-Neural Networks and Deeplearning. -Improving Deep Neural Networks : Hyperarameter tuning, Regularization and Optimization. -Structuring Machine Learning Projects. -Convolutional Neural Networks. -Sequence Models.
May 2020
Coursera
-Introduction to Tensorflow. -Convolutional Neural Networks in Tensorflow. -Sequences, Time Series and Prediction. -Natural Language Processing in Tensorflow.
Feb 2020
Udemy
-Artificial Neural Networks. -Convolutional Neural Networks. -Recurent Neural Networks. -Self Organizing Maps. -Boltzman Machines. -Auto Encoders.
June 2017
Chennai
-Kaizen Robotics Level 1. -Kaizen Robotics Level 2. -Internet of Things using NodeMCU.
Address
Berlin, Germany
Phone
+358 40 1921255
piyush.pungalia05@gmail.com