Senior AI Engineer

Architected a production agentic brand-visibility platform that measures brand representation across OpenAI, Claude, Gemini, Perplexity, and xAI. Built durable orchestration for 540 parallel provider calls per run with retries and provider-specific budgets.

Designed deterministic visibility, sentiment, and share-of-voice scoring from structured outputs, reducing rerun scoring variance from 11% to under 1%. Built citation-verification and chain-of-verification agents, deployed Cortex research workflows on Bedrock AgentCore and LangGraph, and added Langfuse tracing, cost attribution, and golden-set replay evaluation. These improvements reduced hallucinated citations to 0.6%, improved first-pass answer acceptance to 88%, and reduced report cost by 41%.

Built a LangGraph recruiting-agent platform that routed requests among sourcing, screening, scheduling, and outreach agents. Integrated 24 tools across CRM, calendar, and job-board systems while enforcing human approval before candidate contact.

Developed parallel sourcing and reranking workflows, hybrid candidate retrieval using Azure AI Search, and safe retry mechanisms with idempotency keys and candidate locks. Established LangSmith tracing and trajectory-based regression evaluation, improving match precision from 72% to 91%, reducing recruiter hours per hire by 46%, and preventing duplicate outreach in production.

Built LangGraph analytics agent teams in which planner, query, chart, and validation agents answered business questions under execution guardrails. The validation layer checked every figure against source data and reduced analyst turnaround time by 65% across 20,000 records.

Developed curriculum-planning agents that created learning paths from hybrid search across more than 100,000 courses, including prerequisite validation. Also delivered a CrewAI budget-monitoring system with researcher, analyst, and reviewer agents that streamed results to FastAPI WebSocket dashboards, reducing variance-investigation time by three times.

Architected a tutoring-agent system using LangChain and Flask microservices. Implemented learner mastery modeling, concept planning, exercise retrieval, answer grading, and Llama-powered tutoring dialogue, increasing lesson completion by 28% across 12,000 learners.

Built bilingual Arabic and English sales agents with CRM and calendar tool calls protected by confirmation workflows and deterministic human handoff. Designed a six-agent post-session analysis mesh with replay-based evaluation, reaching 89% agreement with human tutors while delivering 1.4-second conversational latency and substantially lower inference cost.

Bachelor of Science in Computer Science.