About Me
Hey, I'm Pooja, a Machine Learning Engineer specializing in retrieval-first AI systems. I build RAG pipelines, semantic search, and production ML infrastructure that actually scales.
I love taking experimental ideas from Jupyter notebooks to reliable, observable services running in production. My sweet spot is architecting end-to-end systems where retrieval quality, latency, and reliability all matter.
What I Focus On
- 🔍 RAG & Retrieval: Semantic embeddings • Vector databases (Pinecone, FAISS) • Cohere reranking • Custom eval frameworks • Retrieval optimization
- ⚙️ MLOps & Infrastructure: MLflow experiment tracking • Docker/Kubernetes deployments • CI/CD pipelines • Model monitoring & observability • FastAPI microservices
- ☁️ Cloud & Scale: AWS (Bedrock, SageMaker, S3, EMR) • GCP (Vertex AI, Cloud Run) • Distributed training • Production serving
Currently exploring: multi-modal retrieval, fine-tuning diffusion models, and making LLMs work reliably in production.
Building an AI-Powered Search with RAG, OpenAI, and Pinecone
OpenAI · Pinecone · Bedrock · Cohere reranker → measurable relevance gains.
Building a Three-in-One Tweet Intelligence Engine
Emotion classification, hashtag suggestions, popularity prediction in one pipeline.
A(I)YE Chef
Built an ingredient detection pipeline using computer vision to identify food items, then match them to recipes. Containerized with Docker and deployed on GCP Cloud Run for scalable inference.