Vetted Unsupervised Learning Professionals

“Built and deployed a production LLM-powered university support chatbot on Azure using a RAG pipeline, focusing on reducing hallucinations, improving latency, and handling ambiguous queries via confidence checks and clarification prompts. Also has hands-on orchestration experience (Airflow/Azure Data Factory), including hardening a demand-forecasting ingestion workflow with sensors, retries, and automated alerts, and uses a metrics-driven testing/monitoring approach for reliable AI agents.”

“AI/ML engineer who led Impacter AI’s production deployment of a specialized outreach LLM (CharmedLLM) fine-tuned on GPT-4.1, cutting API costs ~40% while boosting outreach effectiveness ~60%. Built the supporting MLOps and data infrastructure (MLflow, Kubernetes, PySpark, Kafka) and has agentic AI experience from University of Dayton, using LangChain + RAG and vector search (Pinecone) to improve reliability and reduce hallucinations.”

“Built and deployed a production NLP sentiment analysis system at Piper Sandler to turn noisy, finance-specific customer feedback into scalable insights. Demonstrates strong end-to-end MLOps: fine-tuning BERT, improving label quality, monitoring for language drift, and automating retraining/deployment with Airflow and Docker (plus Kubeflow exposure).”

“ML/NLP engineer with Industrial IoT experience who built an end-to-end anomaly detection and GenAI explanation system: AWS (S3, PySpark, EC2/Lambda) pipelines feeding dashboards, plus transformer-embedding vector search to connect anomalies to noisy maintenance notes and past events. Demonstrated measurable impact (15% lift in defect detection; ~35% reduction in manual review; 35% fewer preprocessing errors) and strong productionization practices (orchestration, monitoring, rollback, data-quality controls).”

“AI/LLM practitioner who has deployed production chatbots across e-commerce, HRMS, and real estate, focusing on retrieval-first workflows for factual tasks like product and property search. Optimized intent understanding and significantly improved latency by using lightweight embeddings and tuning the inference pipeline on Groq (Llama 3.3), while applying modular orchestration and measurable production evaluation.”