Shashank Shekhar

AI ML Engineer

• Working on Agentic AI Search using a Reinforcement Learning framework.
• Designed an LLM-as-a-judge evaluation pipeline that pre-screens data and surfaces likely failures for targeted human auditing, reducing the review set to about 30% of data and cutting manual review effort by about 70%.

Lead / Founding Engineer

• Spearheaded end-to-end development and delivery of multiple AI/ML projects as Project Lead, covering requirement gathering, system design, deployment, scaling, and maintenance.
• Architected and deployed cloud-native, scalable back-end infrastructure for video analytics applications with platform-agnostic principles for on-premises and cloud environments.
• Fine-tuned advanced vision-language models on NVIDIA H100 GPUs and deployed production-grade inference using the vLLM framework for optimized throughput and low-latency serving.
• Configured and optimized Nginx as a reverse proxy and load balancer, including request routing, caching, and security enhancements for reliability under high traffic.
• Designed and developed RESTful APIs using FastAPI with error management, request validation, authentication, and authorization for secure and scalable client-server communication.
• Implemented an agentic back-end pipeline to support horizontal scalability and distributed workloads for enterprise AI automation workflows.

Senior ML Engineer

• Deployed a lightweight object detection algorithm optimized for the Amlogic Neural Hardware Accelerator for real-time performance on constrained hardware.
• Designed and integrated a novel camera tampering detection algorithm using the OpenCV C-API, improving system reliability and security.
• Contributed to simplifying, refactoring, and debugging a legacy C codebase of more than 10,000 lines, improving readability, maintainability, and execution performance.

ML Engineer

• Developed an unsupervised anomaly detection algorithm to identify irregularities and detect fault triggers in ambient light sensor data.
• Experimented with sensor fusion algorithms.

Computer Vision Engineer

• Implemented and deployed computer vision algorithms for real-time detection and tracking using Kalman filters on edge devices, with an optimized detection model size of only 200 KB.
• Managed scientific research and engineering execution for machine learning projects, ensuring both innovation and production-level reliability.

• Conducted pre-training and fine-tuning of foundation models spanning LLMs, text-to-image models, TTS, STT, and vision-language models.
• Adapted pre-trained LLMs to specific domains and tasks using targeted datasets and optimized them for real-world applications.
• Designed and implemented Retrieval-Augmented Generation systems integrating LLMs with company knowledge bases to improve response accuracy.
• Optimized LLM architectures, hyperparameters, and training workflows to improve efficiency, accuracy, and inference performance.
• Developed LLM-powered applications for summarization, translation, question answering, chatbots, and conversational AI.
• Deployed LLMs across Azure, GCP, and on-premises infrastructure.
• Monitored model outputs, corrected biases and errors, and retrained models to maintain high-quality performance.
• Containerized LLM applications using Docker for portability, reproducibility, and scalability.

RL Intern

• Successfully tested Deep Reinforcement Learning algorithms on a self-driving toy car and analyzed Optical Flow and SLAM.
• Led experimentation with RL algorithms including DQN variants, Vanilla Policy Gradient, PPO, and TRPO in OpenAI Gym environments across discrete and continuous action spaces.