Anant Singh Email & Phone Number

Cupertino, California, Us

Improving Document AI systems using computer vision and natural language processing techniques at Applied Machine Learning team.

New York, Us

· Developed LLM (OpenAI GPT4) based RAG generative system for automated and easier Data Analytics. Leveraged embeddings from HuggingFace and OpenAI, alongside LangChain and Chroma Vector DB for RAG system Created a robust REST API using FASTapi and successfully deployed it on AWS.· Enhanced the quality and precision of open-source models such as Llama2 13B, 70B, CodeLlama 13B, and 34B, as well as proprietary models for SQL-Text generation. Employed techniques like Supervised Fine-tuning (SFT), Parameter Efficient Fine-tuning (PEFT) and LORA, thereby ensuring optimal performance and accuracy of the system.· Proficiently utilized a comprehensive technology stack, including Python, PyTorch, HuggingFace, LangChain, FASTapi, MLflow, AWS, and SQL to host, develop, and maintain the Generative AI system, with strong commitment to staying up-to-date on AI.

Eindhoven, North Brabant, Nl

· Designed synthetic data generation pipeline that dramatically decreased data curation expenses down to 20% and minimized computational resources required for training perception models. Used TensorBoard and Optuna to optimize the models.· Trained ViT based Context Autoencoder (CAE) model on MMSelfSup backend and overcame data scarcity for perception models with Supervised Contrastive Learning (SCL). Built models for North America and Europe with more than 1000 classes.

New York, Ny, Us

Projects: - Measuring Personality of Large Language Models-- Assessed whether LLMs have developed a personality and whether we have the right tools to measure personality. Introduced Option-Order Symmetry property necessary condition for realizable self-assessment personality tests. -- Found that many LLMs' personality test responses indicate a lack of consistency in their behavior. Also identify inherent biases in LLMs, which contribute to the unreliability of self-assessment tests. Worked with GPTNeo, GPT2 and OPT.- Multilingual Speech Emotion Recognition System-- Studying Speech models like HUBERT, Wav2Vec for Speech Emotion Recognition (SER) task on 7 different languages. Developing an exhaustive benchmark for SER with support of all the feature extractors and languages available.-- Researching best finetuning strategy for SER feature extractors and ASR models, on Huggingface backend.- Robustness of Semi Supervised Learning Algos in NLP-- Testing robustness of Semi supervised learning algorithms when subjected to out-of-domain noise in unlabeled dataset. Working on UDA, Mixtext, Self- Training and iPET. Worked with BERT and RoBERTa-large.

Eindhoven, North Brabant, Nl

· Developed, trained, and deployed multistage models for Road Sign Markings Detection and Classification in satellite images for HERE Maps. Tuned the models (ResNet, YOLOv3) for optimized performance to work on huge image and GeoJSON datasets.· Improved Classification pipeline by applying super resolution to the satellite imagery for better feature extraction from low resolution images. Worked in Agile development environment, developed with PyTorch and deployed on AWS cloud.

Brooklyn, New York, Us

Contributed to the research team focused at building an application that takes in historical data of the instruments and forecast the return by machine learning and time series methods. Then, apply the aggregated pipeline onto the portfolio and rebalance the allocation regularly. The goal of the project was to establish an application that can perform real-time rebalance portfolio allocation for investors.

Mumbai, Maharashtra, In

· Developed and maintained Machine Learning models for a range of client projects, leveraging Python, Scikit-Learn and AWS· Collaborated with cross-functional team to ensure the successful delivery of data-driven solutions and meet client expectations

· Cofounded a startup focused on providing space saving furniture in Indian metro cities. Started from 2 and grew to 50 in 9 months during the pandemic. Built the website (www.kursistore.com) for the startup.· Managed Data Science team and increased sales by deploying Machine Learning pipelines to predict best marketing channel and budget for a given product to better utilize resources for product marketing.

Ghaziabad, Uttar Pradesh, In

· Assisted Prof from Electronics and Communication Department in analog electronics research. Studied, simulated, and developed research papers on novel analog electronics circuits for publication.· Published: An Active-C Realization for Simulating Electronically Controllable Lossy Grounded Inductance (IEEE)· Published: 5G: An overview of Channels characteristics and modelling techniques (IEEE)

Master Of Science In Computer Engineering, Concentration In Data Science

Bachelor Of Technology, Electronics And Communication Engineering