Muhammad Ehsan Ul Haq Email & Phone Number

Texas, United States

In his current position as an NLP Engineer at OnEngine.ai, a pioneering company specializing in multi-agent conversational systems, he is primarily tasked with the development and enhancement of sophisticated AI-driven dialogue systems. This role entails a deep dive into natural language processing techniques and machine learning algorithms to improve the efficiency and effectiveness of conversational agents. His responsibilities also extend to collaborating closely with cross-functional teams… Show more In his current position as an NLP Engineer at OnEngine.ai, a pioneering company specializing in multi-agent conversational systems, he is primarily tasked with the development and enhancement of sophisticated AI-driven dialogue systems. This role entails a deep dive into natural language processing techniques and machine learning algorithms to improve the efficiency and effectiveness of conversational agents. His responsibilities also extend to collaborating closely with cross-functional teams to integrate these systems seamlessly into the broader product ecosystem. Additionally, he is actively engaged in the continuous exploration of cutting-edge research within the NLP field to identify and implement novel approaches that can elevate the capabilities of OnEngine.ai's conversational platforms, ensuring they remain at the forefront of AI technology. Show less

San Francisco Bay Area

Something super exciting, can't tell more.

San Francisco Bay Area

In his current role as the Head of AI, Ehsan is responsible for maintaining the existing product as well as building new ones. This role also involves a lot of infrastructure management, such as creating failure-tolerant infrastructure and allocating GPU machines to team members. Ehsan is also involved in researching new AI papers and implementing them.

San Francisco Bay Area

As a Tech Lead, Ehsan was responsible for streamlining processes and leading a team of 2-3 developers on a mediating product. He mentored and designed microservice architectures, handled containerization, and deployment of applications.

San Francisco Bay Area

As an Applied Deep Learning Engineer, Ehsan primarily worked on text modelling, including training word embeddings for applications like text comparison and query expansion. He also worked on keyword searching using Elasticsearch and semantic searching using Faiss. In addition to this, he also worked on image generation (stable diffusion, GANs), speech to text, and image to video (first-order motion models).

United States

Ehsan worked as a Machine Learning Researcher in Drug Discovery. The problem was to develop a method that generates new molecules while providing control over the generation process. The task required encoding sequential data (SMILES strings) into a meaningful latent space, with parameters such as specificity and transforming interpolated points within the latent space back to real molecules. The challenge was the high dimensionality of sequential data and the need to generate longer sequences… Show more Ehsan worked as a Machine Learning Researcher in Drug Discovery. The problem was to develop a method that generates new molecules while providing control over the generation process. The task required encoding sequential data (SMILES strings) into a meaningful latent space, with parameters such as specificity and transforming interpolated points within the latent space back to real molecules. The challenge was the high dimensionality of sequential data and the need to generate longer sequences without using attention mechanisms, which typically require feeding the encoded input sequence to the decoder at the time of generation. Ehsan solved the problem by introducing a novel architecture that incorporated a CNN-based VAE into a traditional attention-based sequence-to-sequence model. The VAE was designed to summarize the signal (encoded input) to enable sampling in the latent space, and to reconstruct the encoded input during the attention mechanism. Show less

As a Computer Vision Engineer, Ehsan was responsible for building an end-to-end Live Video Streaming analytics system. The system included data acquisition using streaming protocols like HTTP stream and Dash protocol. A dashboard was designed to stream videos from multiple TV channels simultaneously. The frames from each channel were then fed to computer vision models that carried out real-time, unsupervised face detection and recognition. The analytics dashboard generated statistics such as… Show more As a Computer Vision Engineer, Ehsan was responsible for building an end-to-end Live Video Streaming analytics system. The system included data acquisition using streaming protocols like HTTP stream and Dash protocol. A dashboard was designed to stream videos from multiple TV channels simultaneously. The frames from each channel were then fed to computer vision models that carried out real-time, unsupervised face detection and recognition. The analytics dashboard generated statistics such as who appeared on what TV channel at what time using Dash. This problem was at the intersection of Big Data and Large scale deep learning, where the data not only had high volume, but also high velocity. With each live stream consisting of 5-second video chunks and 24 frames per second, Ehsan worked on a system that could handle large amounts of data and provide valuable insights in real-time. Show less

Pakistan

Ehsan, during his participation in Google Summer of Code, had the opportunity to work with ML4SCI, a part of CERN that employs machine learning to assist researchers in the field of High Energy Physics. His task was to implement an anomaly detection model that would help identify events lying in the region of interest by modeling the background distribution.Ehsan contributed to various aspects of this project, including working on the Data Pipeline to randomize a large dataset using… Show more Ehsan, during his participation in Google Summer of Code, had the opportunity to work with ML4SCI, a part of CERN that employs machine learning to assist researchers in the field of High Energy Physics. His task was to implement an anomaly detection model that would help identify events lying in the region of interest by modeling the background distribution.Ehsan contributed to various aspects of this project, including working on the Data Pipeline to randomize a large dataset using hdf5 format, which helped the models generalize better. Additionally, he modeled the data distributions using Normalizing Flows, contributing to the overall success of the project. Show less

Islamabad, Islāmābād, Pakistan

At GRC SEECS NUST, Ehsan worked as a Computer Vision Researcher where he was involved in multiple projects. Initially, he worked on Medical Imaging where the task was to classify images with cancer indications. Ehsan used Graham Matrix, which captured the texture information of the pathological slide images. He found that texture information was more important than edges and other geometric features while detecting cancer.In his next project, Ehsan worked on Image Segmentation. The goal was… Show more At GRC SEECS NUST, Ehsan worked as a Computer Vision Researcher where he was involved in multiple projects. Initially, he worked on Medical Imaging where the task was to classify images with cancer indications. Ehsan used Graham Matrix, which captured the texture information of the pathological slide images. He found that texture information was more important than edges and other geometric features while detecting cancer.In his next project, Ehsan worked on Image Segmentation. The goal was to identify the precise location of cancerous tumors within images. To accomplish this task, Ehsan employed methods like DeepLab v3. With his experience and knowledge of computer vision techniques, Ehsan was able to develop a highly accurate and reliable system for identifying cancerous tumors in medical images.Apart from Medical Imaging, Ehsan also worked on aerial Imaging. The task was to detect and recognize various types of vehicles, including airplanes, cars, and trucks, from satellite imagery. Ehsan used state-of-the-art techniques in computer vision, including object detection and image classification, to develop a highly accurate system for recognizing vehicles in aerial images. Show less

Islamabad, Islāmābād, Pakistan

Ehsan worked as a Machine Learning Research Intern and has significant experience in developing computer vision and deep learning models. During his internship, he worked on several projects including Person Re-Identification in a multi-camera setting using ResNet for feature extraction and Sparse Subspace Clustering. He also worked on Time Series classification using sequential models like LSTMs, GRUs, and RNNs, and convolutional neural networks for identifying individuals based on their gait… Show more Ehsan worked as a Machine Learning Research Intern and has significant experience in developing computer vision and deep learning models. During his internship, he worked on several projects including Person Re-Identification in a multi-camera setting using ResNet for feature extraction and Sparse Subspace Clustering. He also worked on Time Series classification using sequential models like LSTMs, GRUs, and RNNs, and convolutional neural networks for identifying individuals based on their gait recorded from inertial sensors on cellphones. His work resulted in a published paper in a 3+ impact factor journal, demonstrating his strong research and analytical skills. Show less

Masters, Computer Science

Nano Degree, Deep Learning

Activities and Societies: Sentiment Analysis, Image Classification, Image Generation, Model Deployment One of the 100 recipients.

Nano Degree, Edge Ai For Iot

Activities and Societies: People Counter App Using CCTV Camera, Queue Ordering using CCTV and Finally Mouse Pointer using Eye Gazing. One.

Bese, Computer Software Engineering