Avijit Kundal

As the most senior member of the organization, I led Unscript's AI team, providing strategic direction and overseeing key initiatives:

• Designed custom Lipsync (talking head generation) models based on the StyleGan architecture, incorporating NeRF and 3DMM-based methods, achieving superior Sync and FID scores compared to other state-of-the-art models.
• Experimented with diffusion-based models like DDPM (Denoising Diffusion Probabilistic Models) and DDIM (Denoising Diffusion Implicit Models) for applications in audio cloning and face generation.
• Engineered Video Super Resolution (VSR) pipelines using GFPGAN and GPEN to enhance DeepFake video quality generated from Lipsync models.
• Established Visual Speech Recognition pipelines capable of generating text from lip movements, showcasing a Word Error Rate (WER) comparable to state-of-the-art models.
• Developed and tested multilingual Text-To-Speech (TTS) models for both single speaker and multi-speaker scenarios for voice cloning.
• Optimized existing TTS models, significantly reducing the data requirements for training using advanced techniques.
• Contributed to the design and implementation of generative algorithms spanning Text to Speech, Text to Video, Image to Video, Digital Avatars, and Video Chatbots.

• Spearheaded the company's Data Science strategy for over 10 projects, converting business objectives into actionable Data Science frameworks.
• Led the development and training of advanced Reinforcement Learning Agents (DQN, PPO, SAC) for complex real-world environments.
• Established ML pipelines on AWS, Azure, and GCP, managing data from diverse sources, including IoT sensors.
• Engineered efficient ML tools to accelerate Data Science experimentation.
• Automated Ammonia Recovery and Absorption processes at GHCL using cutting-edge techniques, including DNNs and RNNs.
• Implemented real-time controls for HVAC systems at Unilever Mumbai, resulting in a remarkable 52% reduction in power consumption.
• Managed a high-performing team of data scientists and domain experts for the successful completion of large-scale projects.

• Achieved a 70%+ accuracy in SKU detection for 50+ Reckitt & Benckiser products in low-quality store images through deep learning models.
• Conducted artificial data generation for diverse SKUs using Blender 3D renders, enhancing model robustness.
• Pioneered the design, development, and deployment of a Tooth detection and numbering system for panoramic and periapical X-ray images, utilizing a custom X-ray dental dataset.
• Innovated and implemented a Foot size measurement app, calculating foot size from top and side perspective foot images, demonstrating a holistic approach to image-based applications.

• Innovated and patented Video Vectorization technology, a groundbreaking solution converting pixel-based animated video content into a vector-based format. Implemented using OpenCV and RCNN architectures.
• Pioneered a patented 3D Region Growing algorithm, seamlessly creating 3D segmentations of images across frames through a fusion of image registration and segmentation techniques.
• Authored GPU-accelerated code for diverse Computer Vision tasks, demonstrating proficiency in optimizing algorithms for efficient processing.
• Leveraged extensive experience in running complex algorithms on both local and cloud-based distributed systems, particularly utilizing Ray on Kubernetes.
• Conducted in-depth research aimed at enhancing segmentation and object detection in low-quality images, showcasing a commitment to addressing challenges in product development.
• Devoted significant time to reviewing research papers, ensuring a comprehensive understanding and effective application of cutting-edge solutions in product development.

• Developed a Crop Classifier using CNNs trained on multidimensional satellite imagery of Onion, Potato, Tomato, Chickpea, and Okra fields.
• Generated labeled datasets from unlabelled vector data to identify the locations of each crop.
• Implemented code for detecting ground control points (GCP) in drone images for 3D mapping using object detection and SIFT.
• Preprocessed raw spectral data for training and testing in a CNN.
• Designed a Crop Health Monitoring system by applying a GPU-accelerated KMeans algorithm on time series data of spectral vegetation indices derived from 13-dimensional Satellite Imagery.

• Implemented CNNs and NLP for content filtering on various Facebook pages, enhancing recommendation systems.
• Managed multiple Facebook pages, achieving a total of 400,000 followers within a span of less than 2 years.
• Authored over 20 Science and Technology articles for the engaged followers.