RaviShankar Prasad
About
Detail
Delhi, India
Résumé
Jobs
verified_user
0% verified
- VChief Operating OfficerVocab.Al Pvt. Ltd.Aug 2023 - Mar 2026 (2 years 8 months)• Architected and deployed in-house speaker diarization and ASR systems handling high-volume telephonic speech with custom vocabulary support. • Built end-to-end conversational AI systems in Indian languages (Hindi, and others), integrating custom LLMs with agentic analytics frameworks. • Led fine-tuning of large language models (LLMs) on client-specific domain data using SFT and RLHF techniques. • Designed agentic workflows for automated call-center QA, delivering prescriptive analytics on quality adherence. • Managed cross-functional R&D and product teams; oversaw cloud-based (GCP/AWS) CI/CD production pipelines.
- WScientific OfficerWhispp Inc.Jul 2022 - Jul 2023 (1 year 1 month)• Developed AI-powered models for pathological speech enhancement, whispered speech intelligibility improvement, and noise suppression. • Applied advanced signal processing and deep learning techniques to assist individuals with voice disorders.
Research CollaboratorAalto UniversityJan 2022 - Jun 2022 (6 months)• Classified pathological speech in Alzheimer's patients using acoustic feature analysis. • Investigated voice quality markers and formant structure in pathological speech conditions. • Applied data engineering principles to preprocess, prune, and maintain databases of audio and biomedical data, ensuring high-quality datasets for research analysis. • Collaborated closely with Natural Language Understanding (NLU) researchers on the ideation and development of vector databases, drawing upon knowledge of architectures such as FAISS, Pinecone, and Weaviate to understand their creation and update mechanisms.- IPostdoctoral ResearcherIdiap Research InstituteNov 2018 - Dec 2021 (3 years 2 months)• Conducted multimodal research jointly characterizing speech, respiration, and ECG signals. • Published studies on speech changes due to COVID-19 infection, fetal heartbeat analysis, and marmoset soundscape analysis. • Developed and maintained robust data processing pipelines for physiological data, encompassing data acquisition, pruning, preprocessing, and database management, directly supporting the research objectives. • Ensured model integrity and reproducibility by implementing version control for models and managing relevant APIs, aligning with MLOps principles to streamline the research workflow.
Education
verified_user
0% verified
- Ph.D., Speech Signal ProcessingIIIT HyderabadJan 2014 - Aug 2019 (5 years 8 months)Thesis: Analysis of Dynamics of Vocal Tract System using Zero-Time Windowing (ZTW) Focus: Speech production mechanisms, pathological speech analysis and modeling.
- GB.Tech., Electronics & Communication EngineeringGurukula Kangri UniversityJan 2004 - Dec 2008 (5 years)Faculty of Engineering and Technology.
Projects (professional or personal)
verified_user
0% verified
- VRAG-Based Customer Support ChatbotVocabAI,Mar 2025 - Feb 2026 (1 year)• Implemented Retrieval-Augmented Generation (RAG) using FAISS indexing and ChromaDB for fast document retrieval. • Fine-tuned LLMs for domain-specific knowledge injection; deployed via FastAPI for production use. • Achieved significant reduction in query resolution time through semantic search and LLM-driven responses. • Leveraged the development of this RAG-based customer support chatbot to drive significant business value, by automating the handling of customer issues across live channels. This proposition was particularly well-received by our client, a major e-commerce giant, which handles an estimated 8,000-10,000 calls daily, presenting a clear opportunity for efficiency gains over manual interaction management and the inherent ambigu
- VCall-Center Quality Analytics EngineVocabAI,Oct 2024 - Feb 2026 (1 year 5 months)• Developed a multi-language conversation analytics system for Indian-language call centers. • Generated subjective and objective QA metrics; derived prescriptive insights to improve agent performance. • Reduced manual QA overhead by automating assessment against configurable quality parameters.
- VSpeech-Based Conversational AI FrameworkVocabAI,Aug 2024 - Dec 2025 (1 year 5 months)• Built custom ASR pipelines for telephonic speech with OOV domain-specific word handling via fine-tuning. • Integrated client-specific LLMs with agentic reasoning and TTS for a full-stack voice assistant. • Deployed as a scalable FastAPI microservice on cloud infrastructure. • Leveraged PyTorch library to implement deep convolutional networks for deriving speech input representations, enhancing the framework's ability to process and understand complex audio data. • Utilized scikit-learn for dimensionality reduction and model selection, optimizing the efficiency and performance of the speech-based conversational AI framework.