Digvijay S. Email & Phone Number

San Jose, Ca, Us

Mountain View, California, United States

Collaborating cross-functionally to drive continuous improvement in AI model performance, focusing on reducing hallucinations, improving factuality, and enhancing overall system reliability in a fast-paced startup environment, throughout the seed to Series A journey. Architecting and implementing cutting-edge AI solutions for sales intelligence, leveraging Generative AI, LLMs, and advanced NLP techniques, for 1000s of users across 100s of organizations with industry-defining accuracy.Leading the technical implementation of an end-to-end meeting intelligence platform, integrating various ML models and services into a cohesive system. Utilize a modern tech stack including Python, FastAPI, MongoDB, and AWS to build scalable and robust infrastructure.Developed innovative meeting summarization modules using domain-specific fine-tuning and prompt engineering to extract key outcomes, next steps, and personalized insights.Engineered a sophisticated RAG pipeline for deal intelligence, integrating multiple data sources and implementing novel querying mechanisms. Utilized state-of-the-art embedding techniques and recency-biased retrieval to surface highly relevant information from CRM systems and meeting transcripts.Led 0-1 development of innovative AI email feature that is a differentiator in a competitive market, which has unprecedented 11% initial adoption rate.

San Mateo, California, United States

Built and improved multiple iterations of core product offering – generative customer service bot that takes in a set of documents and builds a conversational experience around it. Integrated novel UX features by extensively working with LLMs: prompt engineering, model training, fine-tuning and distillation, RLHF, QLoRA etc. to bring in multiple revenue streams.Developed and implemented multiple iterations of core IP: novel automated conversational flow discovery algorithms, by solving complex engineering problems to discover conversational flows from conversation datasets.Developed and improved novel conversational flow evaluation method to reduce running time.Brought generative AI features to core product offering to improve user experience.Built data pipeline to add and perform data processing steps (classification, entity tagging, audio transcription etc.) in an automated manner using airflow. Significantly reduced overhead of processing user data for downstream tasks.Collaborated with technical and business facing colleagues to solve challenges with LLM-based product development including hallucination detection, factuality dataset creation, annotating specifications for RLHF etc.Project ownership to rapidly evaluate, prototype, and build new feature and product offerings using latest ML and LLM technologies and tools for business development and revenue streams in a startup ecosystem.

Rockville, Maryland, United States

Worked on Natural Language problems, in both deliverable and R&D projects.Gained experience with AWS environment and tools (S3, EC2, KMS, SQS, SageMaker etc.).Integrated technologies like Docker and Spark for scalability.Practiced industry-standard software development practices: git-code versioning and collaboration, unit testing.Leveraged advantages of transformer architectures.Worked on text similarity tasks at various levels of large text documents.Investigated approaches for making NLP models robust against adversarial attacks.Solved task-specific end-to-end document entity redaction, on complex datasets.Worked on explainable deep learning models using SOTA techniques like Integrated Gradients/Captum, gradient visualizations, attention layer visualizations.

Baltimore, Maryland Area

Johns Hopkins Malone Center for Engineering In HealthcareWorking on segmenting phases of cataract surgery videos.Using coordinate labels to train deep architectures for image segmentation.Creating image and text datasets for time series modeling.Working with deep learning models – CNNs, RNNs, LSTMs.Using transfer learning for fine tuning architectures to act as feature extractors.Implementing SOTA models from research papers, exploring techniques to improve model performances, running experiments and evaluating model results.

Baltimore, Maryland Area

Department of Anesthesiology & Critical Care MedicineDeveloped framework to identify physiologically meaningful ICU patient data, and predict clinical outcomes.Mimic-III Matched Waveform Database was used that contained ICU Electronic Health Records of variable length time series vital signs signals (like heart rate, BP, SpO2 etc.), sampled at 0.01667 and 1 Hz.Worked in collaboration with clinical experts to develop artifact correction algorithm, which when applied for data cleaning and prepossessing to the time series data, gave approximately 20,000 physiologically viable ICU stay records belonging to 10,000 patients, for 8 considered signals of interest.Developed predictive analytics algorithms in Python (package) to classify certain transitional and quality improvement-related clinical outcomes of interest using various classification methods.Institute for Data Intensive Engineering and Science (IDIES) 2018 poster presentation: Developing a Framework to Enable Large-scale Analysis of Physiologic and Clinical Data Obtained through Electronic Health Records, Ali Afshar, Digvijay Singh, Aidan Crank

Baltimore, Maryland Area

Department of Applied MathematicsModified PATH algorithm to adopt seeding (seeded PATH or sPATH).Implemented sPATH in C++ using PATH from GraphM package.Ran experiments to compare memory and time complexity with sPATH and SGM.

New Delhi Area, India

Institute for System Studies and AnalysesWorked on war game simulation software.Developed intelligent rule engine in Java (Eclipse IDE).

New Delhi Area, India

Human Computer Interaction project.Developed real-time LED blink detection with Arduino Uno, using NeuroSky Mindwave brainwave technology.Developed blink detection algorithms, implemented in Objective-C using Xcode IDE and Mindwave APIs.

Master'S Degree

Master’S Degree

Bachelor’S Degree

Education record