Shashank Shekhar Pandey

• Architected and deployed a production-grade real-time Voice AI platform integrating Deepgram ASR, GPT-4o Mini, Plivo Telephony, Piper TTS, LangGraph, and WebSockets for appointment scheduling, order tracking, and customer support workflows.
• Engineered bidirectional streaming speech pipelines with conversational memory and state-driven workflow orchestration, enabling natural multi-turn interactions with sub-second response latency.
• Implemented LangGraph-based workflow management, tool invocation, and context retention mechanisms to improve dialogue continuity and task completion rates.
• Designed and built an automated call auditing platform leveraging Amazon S3, AWS Transcribe, and LLM-based evaluation workflows — reducing manual QA effort by 65% and review turnaround time by 70%.
• Designed and scaled a hybrid RAG architecture across 1,200+ enterprise documents using BGE embeddings, BM25 retrieval, Reciprocal Rank Fusion, and cross-encoder reranking.
• Reduced GPU memory requirements by 60% through deployment of 4-bit and 8-bit quantized inference pipelines using bitsandbytes on AWS EC2.
• Fine-tuned open-source LLMs using LoRA and QLoRA with PEFT on domain-specific datasets, achieving a measured 28% reduction in hallucination rates through structured benchmarking.

• Reduced end-to-end LLM response latency by 40% through LangChain workflow optimization, retrieval pipeline improvements, and intelligent caching strategies.
• Architected StoryTime AI using multi-model orchestration across LLaMA, Groq, and GPT-family models with task-specific routing logic, improving user engagement by 35% and narrative consistency by 20%.
• Built AI-powered parental guidance systems using FastAPI, LangGraph orchestration, custom tools, and LangSmith observability, significantly improving workflow autonomy and debugging efficiency.
• Developed multi-modal AI services integrating REST APIs, WebSockets, and TTS pipelines, increasing recommendation accuracy by 25%.

• Built and deployed a LLaMA-based NL-to-SQL platform enabling business users to query enterprise data through natural language, eliminating manual reporting workflows.
• Reduced forecasting MAPE by 30% through development of a Prophet-based demand forecasting solution deployed on GCP Vertex AI.
• Improved marketing campaign ROI by 25% through predictive modeling using Snowflake-based feature engineering, LightGBM training pipelines, and AWS SageMaker deployment.
• Increased revenue per user by 15% through customer segmentation and behavioral analytics models delivered via automated Docker-based CI/CD workflows.

• Developed and deployed revenue forecasting and churn prediction models using Random Forest, XGBoost, Bayesian optimization, and grid-search tuning techniques.
• Built robust preprocessing pipelines including outlier handling, feature engineering, categorical encoding, and missing-value treatment — reducing pipeline failures by 40%.
• Delivered Tableau and Power BI dashboards adopted by business leadership for operational reporting and strategic decision-making.

Founded Quantora Analytics to deliver end-to-end AI consulting — from strategy and architecture to deployment and MLOps. Leads architecture, client delivery, the Quantora AI Academy internship program, and builds production AI systems for enterprise clients across India and globally.