Vetted Reinforcement Learning Professionals

“Built and deployed a fine-tuned Qwen 2.5 14B model into production at Dextr.ai as the backbone for hotel-operations agentic workflows, running on AWS EKS with Triton and TensorRT-LLM. Demonstrates strong cost-aware LLM engineering (QLoRA, FP8/BF16 on H100) plus rigorous benchmarking/observability (Prometheus, LangSmith) with reported sub-30ms TTNT. Previously handled long-running ETL orchestration with Airflow at GE Healthcare and Lowe's.”

“Robotics perception engineer focused on autonomous driving 3D detection, integrating PETR embeddings into BEVFormer and tackling hard orientation/temporal alignment issues in multi-camera BEV pipelines. Uses Gazebo with custom sensor plugins to validate calibration, timing, and transforms, and blends synthetic labels with real imagery for scalable 3D box generation.”

“AI engineer who deployed a production LLM-powered safety system for an education platform, combining rule-based checks, multi-LLM verification, and selective context (prompt+image vs image-only) to prevent explicit prompts/images from getting through. Strong focus on reliability via benchmarking, trace-based failure analysis, and continuous improvement driven by stakeholder feedback and manual review.”

“Built an AI basketball shooting coach that analyzes player form against NBA players and recruited 30+ beta users via Reddit to drive iterative UI/workflow improvements. Also has internship experience building an administrative server and coordinating API/database compatibility with another client server, emphasizing communication and integration quality.”

“AI/ML engineer with healthcare domain depth who led a HIPAA-compliant, production LLM system at McKesson to automate clinical document understanding—extracting entities, summarizing provider notes, and supporting authorization decisions. Hands-on across Spark/Python ETL, Hugging Face + LoRA/QLoRA fine-tuning, RAG, and cloud-native MLOps (Airflow/Kubernetes/Step Functions, MLflow, blue-green on EKS/GKE), with explicit work on PHI handling and hallucination reduction.”

“Machine Learning Engineer currently at Webster Bank building an enterprise-scale LLM agent for Temenos Journey Manager/Maestro, using RAG-style multi-stage retrieval with FAISS/Pinecone, hybrid dense+sparse search, and LoRA fine-tuning optimized via NDCG/MAP and A/B testing. Previously handled messy incident/telemetry data at Deuta Werke GmbH with deterministic + fuzzy entity resolution, and has strong production data engineering experience across Spark/Hadoop and Python ETL systems.”

“Built and scaled a multi-agent LLM automation pipeline during a fintech internship, growing from a rapid 1-week proof-of-concept to a 15+ agent hierarchical system that cut market brief report generation time from ~5 hours to under 30 minutes. Hands-on with agent frameworks (Haystack, CrewAI, LangChain) and experienced in debugging agent communication issues via sandboxed modular testing and context/token management; also regularly gives architecture-first technical demos at multiple hackathons and university events.”

“AI/LLM engineer who built a production autonomous GenAI content ecosystem that generates short-form scripts, extracts viral highlights from long-form video, and dubs content into 33+ languages. Focused on making LLM outputs production-safe via schema enforcement, token-to-time alignment, critic-agent verification, and scalable async orchestration—cutting manual workflows by ~90% and saving $200k+ annually.”

“Built and deployed a production long-form article summarization system using BART/T5/PEGASUS, tackling real-world constraints like token limits, latency/quality tradeoffs, and factual drift via chunking/merge logic and constrained decoding. Uses pragmatic Python-based pipeline orchestration (scheduled jobs, modular scripts, logging/retries) and iterates with stakeholder feedback to make outputs genuinely useful for content workflows.”

“Robotics software candidate with experience spanning simulation (MuJoCo, Gazebo, Webots) and ROS1/ROS2 development, including hardware-oriented work on a hexapod and a Mecademic Meca500 R3 arm. Built a visually guided interactive indoor robot system using a CV pipeline plus POMDP + imitation learning with PPO-based residual RL, and has practical debugging experience improving LiDAR SLAM stability and migrating sensor interfaces from ROS1 to ROS2.”

“Built and deployed a production agentic LLM knowledge assistant that answers complex questions over internal documents, APIs, and databases using a RAG architecture (FAISS/Pinecone) and LangChain/LangGraph orchestration. Emphasizes production-grade reliability and hallucination control through grounding, confidence thresholds, validation, retries/fallbacks, and full observability (logging/metrics/traces) with continuous evaluation and feedback loops.”

“Built a production-oriented multimodal RAG "Fix Assistant" with FastAPI, Tavily search, BM25 + cross-encoder reranking, and a local Phi-3.5 model, emphasizing strict grounding and fallback/verification modes to prevent hallucinations. Also has hands-on federated learning experience using STADLE to orchestrate edge-node training and aggregation for EV telemetry data, plus experience communicating AI results to non-technical stakeholders (traffic RL/congestion outcomes).”

“Backend/ML engineer working on a battery energy storage system operations dashboard: built a Flask backend integrated with OAuth and a separate FastAPI optimization/simulation service, deployed via Docker CI/CD to Azure Container Apps. Strong in productionizing ML (AzureML to batch endpoints) and in performance/scalability patterns (Postgres indexing/JSONB, per-unit data isolation, async throttling + caching for year-long CPU-intensive simulations across 40+ scenarios).”

“Built and deployed LLM-powered agentic systems including a multi-agent travel planning assistant using LangChain, RAG (FAISS), real-time APIs, and a supervisor agent to manage coordination and reduce hallucinations. Also developed a Text-to-SQL system with schema-aware validation guardrails, and collaborated with drilling domain experts at CNPC USA to build an ML model predicting rate of penetration (ROP).”

“Backend/ML-leaning engineer who built a content-based event recommender for FlowMingle using embeddings + HNSW vector search on Google Cloud, with Firebase as the backend and a managed recommendation lifecycle (15 recs/user, daily async generation, weekly deletion) now serving 1500+ users. Also led a cost-driven migration of ConvAI services to Azure AI using parallel request testing from a Unity client, with post-migration monitoring via logs and model evals; contributed to a Massachusetts law-enforcement conversation analysis system by expanding ingestion to PDF/TXT/Excel and multi-file inputs.”

“Robotics/AV engineer with strong ROS2 and autonomy stack integration experience, including bringing Autoware Universe up on a real Lexus autonomous vehicle platform. Also built a hierarchical reinforcement learning proof-of-concept for Boston Dynamics Spot (navigation + manipulation) and tackled sim-to-real challenges by implementing PD torque conversion for Jetson-based hardware; improved localization accuracy via GNSS+EKF fusion with a reported 28% drift reduction.”

“Robotics software researcher who led a Boston Dynamics SPOT project on non-prehensile manipulation of heavy boxes, combining MuJoCo-based RL, ViT-based perception, and SPOT SDK control; the work is under review for ICRA 2026. Also built a ROS planning-and-learning stack on a LoCoBot using PDDL task planning, RTAB-Map SLAM, MoveIt motion planning, and RL to recover from execution failures.”

“LLM/agent engineer who built a production LangGraph multi-agent orchestrator connecting GitHub and APM/observability signals with a chain-of-verification loop for root-cause analysis. Emphasizes pragmatic architecture (start simple with state summaries), performance tuning (async LLM calls, Docker), and rigorous evaluation (LLM-as-judge, adversarial testing, hallucination/instruction adherence metrics, tool-call tracing) while iterating with non-technical stakeholders via A/B testing.”

“Robotics/ROS2-focused early-career engineer who built a stereo visual-odometry SLAM system for autonomous navigation and optimized it to run reliably in real time on Raspberry Pi. Strong in sensor fusion (camera+IMU), ROS2 debugging/profiling, and distributed robotics/IoT pipelines (ROS2 + MQTT + cloud), with added experience extracting WiFi CSI for sensing/localization and shipping via Docker + GitHub Actions CI/CD.”

“Robotics software engineer focused on ROS2 autonomy, with hands-on work building a monocular visual odometry system on KITTI (including GPS-based scale correction and RViz trajectory visualization) and an end-to-end Gazebo simulation integrating URDF, slam_toolbox, and Nav2. Demonstrates strong practical debugging skills around TF frames, lifecycle nodes, and Gazebo plugin/version compatibility.”

“AI/LLM engineer with recent production experience at UnitedHealth Group building an end-to-end RAG system over structured EMR data and unstructured clinical notes, including evidence retrieval, GPT/LLaMA-based reasoning, and a validation layer for reliability. Strong in orchestration (Kubeflow/Airflow/MLflow), prompt engineering for noisy healthcare text, and rigorous evaluation/monitoring with gold-standard benchmarking, plus close collaboration with clinical operations stakeholders.”

“Controls and automated-vehicle engineer from the EcoCar EV Challenge who led an 8-person team implementing ACC and lane-centering directly on a vehicle (1st place result). Strong in CAN-based debugging, simulation-to-real deployment (Simulink to C++/dSPACE), and distributed robotics communication using DDS, with additional exposure to multi-agent RL and control barrier functions for coordinated driving.”

“At Dreamline AI, built and productionized an AWS-based incentive intelligence platform that uses Llama-2/GPT-4 to extract eligibility rules from unstructured state policy documents into structured JSON, then processes them with Glue/PySpark and serves results via Lambda/SageMaker/API Gateway. Designed state-specific ingestion connectors plus schema validation and automated checks/alerts to handle frequent policy/format changes without breaking the pipeline, and partnered with business/analytics stakeholders to deliver interpretable eligibility decisions via explanations and dashboards.”

“ML/LLM engineer at Charles Schwab who built a production loan-advisor chatbot integrated with internal knowledge and loan-calculator APIs, adding strict numeric validation to prevent rate hallucinations and optimizing context to control costs. Also runs ~40 Airflow DAGs orchestrating retraining/ETL/drift monitoring with an automated Snowflake→SageMaker→auto-deploy pipeline, and uses rigorous testing plus canary rollouts tied to business metrics and compliance constraints.”