Vetted Model Fine-tuning Professionals

“AI/LLM engineer with healthcare domain experience who built a production clinical support “chart bot” for Molina, including PHI-safe ingestion of 200k+ PDF policies, vector retrieval, and a fine-tuned LLaMA served via vLLM on ECS Fargate. Demonstrated measurable performance wins (HNSW + namespace partitioning; 30% inference latency reduction) and a rigorous evaluation/monitoring approach, while partnering closely with nurses and operations teams to shape workflows and guardrails.”

“Built a production AI-powered sales automation system that reads inbound product enquiry emails, extracts structured data, and routes decisions via a rules-based workflow integrated with a product database. Leverages Gemini structured outputs/schema plus option-based prompting and validation to keep responses reliable, and optimizes latency by breaking agent reasoning into smaller LLM calls; evaluates workflows with LangSmith and metrics like completion rate and accuracy.”

“Built and deployed an LLM-powered mental health therapy assistant at AppHealth that segments users by stress level and delivers personalized, non-medical guidance. Implemented healthcare-focused safety guardrails (secondary LLM output filtering) and a multi-agent router workflow validated via statistical tests and therapist review, then scaled training/inference on AWS (EC2/Lambda/DynamoDB) with Kubernetes.”

“Internship project at Builder Market: built an end-to-end production multimodal LLM application that estimates renovation/replacement costs from appliance photos (CLIP embeddings) or text descriptions, combining fine-tuning with agentic RAG. Focused heavily on real-world performance constraints—latency and cost—using parallel agent workflows, model routing to smaller/open-source models, re-ranking, and retrieval chunking, and collaborated closely with CEO/co-founders to deliver the solution.”

“Built and deployed a production RAG system for financial/compliance teams using GPT-4, Claude, and local models to retrieve and summarize thousands of internal documents with strong security controls (role-based retrieval, PII masking). Drove significant operational gains (30+ hours/week saved, ~35% productivity lift, ~45% faster responses) and orchestrated end-to-end ingestion/embedding/index refresh pipelines with Airflow, S3, and SageMaker while partnering closely with compliance stakeholders on auditability and traceability.”

“Built a production, real-time insurance claims document-understanding and fraud-detection pipeline using TensorFlow + fine-tuned BERT, deployed on AWS (SageMaker/Lambda/API Gateway) with automated retraining via MLflow and Jenkins. Addressed noisy documents and latency using augmentation and model distillation (3x faster), cutting claims ops manual review by ~50% and reducing fraudulent payouts.”

“Built and deployed an AI agent to help patients navigate complex housing information by scraping and normalizing unstructured data across all 50 U.S. states, then layering a LangChain RAG system with MMR re-ranking to reduce hallucinations. Experienced in orchestrating multi-agent workflows (LangGraph/CrewAI) and production reliability practices (Pydantic-validated outputs, LLM-as-judge evals, tracing). Also delivered stakeholder-facing explainability via SHAP dashboards for a loan-approval predictive model at Welspot.”

“AI/ML engineer who led production deployment of a multimodal (text/video/image) RAG system on GCP using Gemini 2.5 + Vertex AI Vector Search, scaling to 10M+ documents with sub-second latency and +40% retrieval accuracy. Strong MLOps/orchestration background (Kubernetes, CI/CD, Airflow, MLflow) with proven impact on reliability (75% fewer incidents) and deployment speed (92% faster), plus experience delivering explainable ML (XGBoost + SHAP + Tableau) to non-technical retail stakeholders.”

“Robotics/perception engineer focused on production-grade, real-time systems—optimized self-supervised segmentation on Jetson Nano from ~6–10 FPS to ~20–25 FPS and scaled experimentation/deployment by unifying 15+ edge models in a modular PyTorch Lightning framework. Experienced integrating distributed LiDAR-camera fusion via gRPC/protobuf into mission planning, migrating ROS1→ROS2 Foxy for multi-drone perception, and adding Prometheus-based observability for long-running deployments.”

“UE5 UI engineer who has shipped production-ready HUD/menu frameworks using C++/Slate/UMG and CommonUI, emphasizing MVVM-style architecture for maintainability and designer-friendly iteration. Strong in UI profiling/optimization (Unreal Insights + Slate Profiler), including Slate list virtualization and event-driven updates that improved UI frame time by ~30% in heavy menu scenarios.”