Built consultant-project matching engine using FastAPI, Go, PostgreSQL, vector embeddings, and custom ML models, serving 50K+ requests/day at sub-100ms latency

Integrated OpenAI and Anthropic APIs to power conversational search, auto-generated consultant profiles, and semantic matching

Developed real-time analytics dashboards and data pipelines for financial insights and client decision-making

Improved retrieval relevance 67% by building a RAG pipeline with PyTorch embeddings and reranking, validated through A/B testing across user cohorts

Cut deployment time 70% by migrating a monolith to AWS Lambda serverless and shipping an ML evaluation framework with regression detection

Built multilingual accessibility features for Arabic and Portuguese students, including speech-to-text and predictive input, deployed to live classrooms

Achieved 99.9% SLA uptime on critical Go microservice by diagnosing OOM crashes and implementing connection pooling, circuit breakers, and retry with exponential backoff

Reduced hallucination rate 40% by building RAG retrieval system with vector database, LLM orchestration, and automated evaluation framework

Shipped REST/gRPC APIs with JWT/OAuth2 auth, deployed on Kubernetes with CI/CD, Prometheus/Grafana observability, and automated rollback

Reduced production defects 30% by owning testing, code reviews, and automated QA across the full SDLC