Machine Learning Engineer at New Relic (2024-07 – Present)
Architected and deployed MLOps systems, LLM evaluation frameworks, and secure agent platforms on AWS.
- Architected an end-to-end MLOps system using Airflow to orchestrate daily evaluation and retraining pipelines, tracking experiments in MLflow and serving the optimal model via a FastAPI inference endpoint.
- Designed a Data Retriever agent with dual-mode APIs (streaming/non-streaming) to optimize payload delivery and implemented an event-driven SQS + Lambda workflow that automates monitoring for metrics, reducing manual triage time by ~15 hours/month.
- Engineered a comprehensive LLM evaluation framework using Airflow that automatically samples and benchmarks 1000+ records per run, ensuring continuous performance tracking for internal LLM and RAG systems.
- Implemented Knowledge Augmented Generation techniques that improved Retrieval Recall@5 by 40% and increased Answer Precision by 25%, significantly reducing hallucination rates in the RAG pipeline.
- Hosted fine-tuned Small Language Models (SLMs) on GPU infrastructure using vLLM, optimizing for high throughput (117 tokens/sec) and handling 30 RPM with 98% reliability while reducing inference costs by ~60% compared to H100-based hosting.
- Architected a secure Bring Your Own Agent platform on AWS Fargate designed to scale to 200+ concurrent agents (currently hosting 30 live), featuring automated vulnerability scanning and trigger-based streaming execution.
Machine Learning Engineer at Outsystems (2023-07 – 2024-07)
Engineered middleware layer, built centralized LLM Gateway, and automated infrastructure provisioning.
- Engineered a middleware layer to capture and structure request/response payloads, building high-quality datasets for future RLHF (Reinforcement Learning from Human Feedback).
- Built a centralized LLM Gateway, serving as the single access point for all internal models to ensure consistent security, rate limiting, and observability.
- Automated infrastructure provisioning using Terraform and optimized CI/CD pipelines, reducing deployment friction and ensuring reproducible environments.
Junior Machine Learning Engineer at Mad Street Den (2022-07 – 2023-07)
Improved computer vision model accuracy and deployed ML solutions at scale.
- Worked on improving the computer vision model accuracy on client data from 80% to 95% which helped in company's growth.
- Used pyspark to fetch client data from delta table and use that data to improve models, dockerize it, deploy it at scale and monitor the traffic in kibana.
