Ali Haidar
Full-Stack AI Engineer | Agents, SaaS & Automation
A better LLM won’t fix a weak retrieval pipeline.
In production RAG systems, answer quality depends heavily on what happens before the prompt reaches the model.
Chunking, embeddings, metadata filters, retrieval, reranking, grounding, permissions, and evaluation all determine...
Background jobs become risky when retrying them can repeat the business action.
In production SaaS systems, workers can crash, messages can be delivered again, APIs can timeout, and jobs can partially complete.
I design asynchronous workflows around idempotency, controlled...
A multi-tenant SaaS platform needs more than a tenant_id.
Reliable tenant isolation has to work across APIs, database queries, background jobs, caching, storage, integrations, and authorization.
I like designing systems where tenant boundaries are enforced by the architecture...
Connecting two APIs is easy. Keeping two systems in sync is the real engineering challenge.
Production integrations have to survive timeouts, duplicate events, partial failures, retries, and records that drift out of sync.
That’s why I design integrations around idempotency,...
A slow SaaS product isn’t always caused by slow code.
Often the real issue is the full request path: inefficient queries, too many round trips, large payloads, duplicate frontend requests, or unnecessary processing.
I approach performance work by measuring each layer first, then...
