Viprav Anand

• AI Leasing Agent: Designed and deployed an AI Leasing Agent using advanced LLMs with Retrieval-Augmented Generation (RAG), prompt engineering, and fine-tuning. Optimized LLM calls to increase customer lead conversion rates from 10% to 40% and achieved a >95% pilot-to-customer conversion rate in B2B pilots.
• LLMOps & Observability: Integrated LangSmith for comprehensive LLMOps, establishing robust observability, monitoring, and evaluation pipelines for production systems.
• LLM Evaluation Frameworks: Developed custom LLM evaluation and “LLM-as-a-Judge” frameworks for scalable and automated assessment of model outputs. Built an iterative cycle of evaluation, error analysis, and model updates to ensure consistent quality and relevance in production.
• Scalable System Architecture: Engineered a low-latency, asynchronous web server architecture using FastAPI, Celery, and Redis. Leveraged async I/O, efficient caching, and distributed background processing to scale the application 50x with no increase in p95 response times, ensuring reliable, low-latency API performance under heavy load.
• AI Voice Bot: Built an AI Voice Bot using Twilio Voice and OpenAI’s MultiModal Realtime API over WebSocket connections, enabling natural, dynamic phone conversations with sub-3-second response latency.
• Team Leadership: Mentored and led teams of junior and mid-level ML engineers; collaborated closely with cross-functional stakeholders to drive successful project delivery.

• Worked on identifying driving behaviour and road segment features using ML/DL techniques

Key Projects

Anomaly Detection in IMU Sensor Data

• Developed a LSTM based Auto-encoder DL model for identifying bad road patches and speed-breakers based on IMU sensor data

Trip Statistics and Events

• Developed algorithms to derive driver trip stats and deduce driving events like Hard Braking based on collision risks
• Worked on data cleaning and smoothening of streaming IMU sensor data

• Worked for ML team at Kruzr on building driver safety platform

Key Projects

Hard Turn and Lane Change Detection

• Developed a model for hard turn detection using Xgboost model based on both time domain and frequency domain features on gyroscope and accelerometer data

Speeding Data Exploration and Analysis

• Used K Means Clustering to analyse speeding events with respect to traffic, weather and IMU sensor data to develop a multi-layered model for performance improvement of existing model

Advised political clients on political strategy through analysis of qualitative and quantitative primary survey data,