Saad A.

Ship production features across a multi-tenant enterprise GenAI platform (RAG, LLM assistants, agentic pipelines, tenant admin) serving regulated industries including financial services and insurance.

Own end-to-end work in the multi-format document ingestion pipeline (PDF, Office, HTML, email, images), covering parsing, chunking, embedding, and vector upsert, running on a worker-thread pool with AMQP-driven scheduling.

Integrated Azure AI Document Intelligence and Azure OpenAI into the ingestion path with multi-endpoint load balancing and a span-based page composer that fuses text, tables, and vision-LLM figure descriptions.

Built agentic Python microservices (FastAPI + TaskIQ + Redis) that enrich ingested content with LLM-generated metadata and image understanding, wired into the Node/NestJS backend via typed adapters and webhooks.

Contributed to the hybrid retrieval layer combining vector search (Qdrant), keyword search (Elasticsearch), and an external reranker, enforcing tenant-scoped access controls in PostgreSQL/Prisma.

Shipped full-stack features across the chat/assistants product, knowledge-base upload app, and tenant admin console, using NestJS + GraphQL + Prisma on the backend and Next.js 14 + Redux Toolkit + Tailwind on the frontend.

Drove performance and reliability work across hot paths (Postgres query tuning, worker concurrency, batching); delivered features behind feature flags with GitOps rollouts (Helm + ArgoCD on Kubernetes).

Utilized the ELK stack (Elasticsearch, Logstash, and Kibana) to analyze Apache server logs for detecting anomalous behavior and potential security issues;

Applied supervised machine learning techniques to classify business text messages by industry, enabling structured insights and automated categorization.