Vinit Singh

Financial Intelligence & Stock Advisory AI Fine-tuned DeepSeek-R1-Distill-Llama-8B and Llama 3B using LoRA to build a Warren Buffett-inspired robo-advisor, delivering stock-specific advice and value investing guidance in Buffett's distinctive style. Developed an investment scoring engine for NSE 500 companies, evaluating profitability, growth, leverage, cash flow, and efficiency metrics to rank listed stocks. Implemented stock news sentiment classification and benchmarked Peter Lynch-trained model responses for comparative investment analysis. Conversational Voice AI & Telecaller Agent Architected an end-to-end AI telecaller pipeline for automated loan recovery — covering training data extraction from call recordings, transcription diarization, and augmentation with simulated conversations. Fine-tuned Ultravox (multimodal LLM) on real loan recovery data with Indic language support, enabling multi-turn topic-wise conversations. Benchmarked TTS models (Kokoro, Parler-TTS-Indic) for Indic language output and optimized latency via vLLM, reducing Hindi response times from 30 seconds to 3.5 seconds. Speech Language Model Research & Evaluation Evaluated open-source speech LMs — LFM2-Audio-1.5B, Qwen2.5-Omni-3B, mini-omni2, and Moshi — identifying Qwen2.5-Omni-3B as the top performer in voice quality. Implemented WebRTC pipelines for real-time audio transfer and latency benchmarking across LFM and Qwen models, with VAD support. NLP Classification & On-Device Deployment Built a BERT-based SetFit model for 3-class text classification and deployed it via ONNX runtime for on-device inferencing within the TVAM Android app. Drove query classification initiatives, producing modular reports that enhanced system intelligence and informed strategic AI decisions. Personality Bot & Audio Processing Led the Simi Garewal Personality Bot project, handling dataset extraction and large-scale speaker diarization using Pyannote.

Designed and delivered three enterprise-grade AI courses on Udemy, reaching thousands of learners across 150+ countries. Course 1: Mastering Voice AI : From ASR to Emotion AI to Voice Cloning: Comprehensive training on ASR, Emotion Detection, Voice Cloning, and speech-language models. Enroll here: https://www.udemy.com/course/mastering-speech-language-models-from-asr-to-emotion-ai/?referralCode=151B17C41EF9FA8F0772 Course 2: Mastering Agentic AI: From Prompt to Protocols to Production: From Prompt to MCP - A2A to Production (new launch): 37+ hours covering advanced prompting (CoT, ReAct, ToT), RAG pipelines with vector databases, multi-agent systems (hierarchies, debate, pub-sub), tool calling & CUAs, HITL workflows, production observability (LangSmith, OpenTelemetry, Prometheus, Jaeger, ELK Stack), and secure deployment with MCP/A2A protocols. Enroll here: https://www.udemy.com/course/mastering-agentic-ai-from-prompt-to-protocols-to-production/?referralCode=C5D805FB0A796752DFD5 Course 3: Mastering Computer Vision: From Pixel to Detection to Gen-CV: From Pixel to Detection to Gen CV — 6 comprehensive modules covering image processing & OpenCV fundamentals, CNNs from first principles, advanced architectures (ResNet, EfficientNet, ViT), object detection (YOLO, Faster R-CNN, DETR), image segmentation (U-Net, Mask R-CNN, SAM), and generative models (VAEs, GANs, Vision Transformers) — built for production-ready, real-world CV development. Enroll here: https://www.udemy.com/course/mastering-computer-vision-from-pixel-to-detection-to-gen-cv/?referralCode=83A04C9BBDD1A2C02B6D Udemy Business Top 3% approved content – focused on production-ready, real-world AI development for engineers, developers, and enterprises.

AI-Powered Creative & Marketing Automation: Consulted for a confidential client building an AI-powered creative campaign and marketing intelligence platform. Explored and implemented LLM orchestration techniques to automate content generation pipelines and streamline campaign production workflows. Contributed to architecting multi-step prompt chains for generating visuals, copy, and marketing strategy — bridging LLM capabilities with real-world creative automation needs.

EdTech AI & Adaptive Learning Platform: Led AI/ML integration and product strategy for BYJU'S Digital, overseeing a platform serving a peak enrollment of 250,000+ students. Spearheaded the implementation of content recommendation engines, adaptive learning algorithms, and personalisation pipelines — tailoring learning experiences to individual student needs at scale. Drove student performance analytics initiatives to surface actionable insights, enabling data-driven decisions across the product roadmap. Managed and aligned cross-functional teams across AI implementation and product development to deliver measurable improvements in learner engagement and platform intelligence.

Computer Vision & Edge AI: Contributed to research and development of Computer Vision solutions deployed on Nvidia edge hardware for enterprise environments. Worked on applied CV applications spanning health & safety monitoring, product quality management, guest experience & crowd monitoring, and inventory management — targeting manufacturing and warehousing use cases. Gained hands-on experience in edge AI deployment, real-time video inference pipelines, and enterprise-grade computer vision system integration.