Vetted PyTorch Professionals

“Built a web-based interface that connects an internal bug system to an LLM for initial debugging and issue classification, aiming to boost QA and software engineer efficiency while balancing latency and accuracy. Worked as a one-person project and managed constraints like limited hardware and difficulty extracting team debugging context, relying on manager communication and rapid modeling to validate direction.”

“Backend engineer (4 years) who built an end-to-end Python backend for a patent-pending in-car massager/heater system, including GraphQL data modeling and Bluetooth integration with an ESP32 microcontroller (reverse engineered a niche protocol). Also has strong platform experience: on-prem Kubernetes/CI-CD (Jenkins/GitLab, exploring ArgoCD GitOps), Terraform-based infra workflows, a RabbitMQ messaging library used across microservices, and an on-prem migration of ~30 critical applications with rollback/parallel-run strategy.”

“Player-coach engineer/lead on a high-scale research integrity platform ("Lighthouse") that flags fraud/manipulation signals across ~3M academic manuscripts per year. Owns architecture decisions (ADRs), implements across Go/Java/React services, and introduced NLP (SciBERT embeddings + human-in-the-loop) to assess out-of-context citations while also handling production incidents with a data-consistency-first approach.”

“Robotics software engineer (master’s student) who placed 3rd in the CMU VLA challenge and presented at IROS, building an LLM-powered language system (Gemini 2.5) for mobile-robot scene Q&A and language-based navigation. Hands-on ROS1/ROS2 experience including ros2_control + PILZ planning for a KUKA arm, plus simulation (Gazebo) and containerized submissions with Docker.”

“Robotics architect intern/new-grad focused on warehouse AMRs, building ROS2 sensor-fusion and SLAM stacks (FastSLAM-style particle filter) and validating in Gazebo with ground-truth metrics. Also interned at ASML debugging real-time in-vacuum robot behavior via Python state-machine telemetry scripts, identifying a firmware driver issue impacting throughput.”

“Backend/retrieval-focused engineer with production experience at Perplexity building a large-scale real-time Q&A system using retrieval-augmented generation, emphasizing low-latency, high-quality answers through ranking, context optimization, and caching. Also has orchestration experience from both product-facing LLM pipelines and large-scale infrastructure workflows at Meta, and has partnered with non-technical stakeholders to align AI trade-offs with business goals.”

“AI/LLM engineer with production experience at NVIDIA, where they fine-tuned and deployed a financial-services chatbot and cut latency ~50% using TensorRT + NVIDIA Triton, scaling via Docker/Kubernetes. Also has consulting experience at Accenture delivering a predictive maintenance solution for a logistics network, bridging non-technical stakeholders with actionable dashboards.”

“Backend/infra-focused engineer with experience spanning Go-based MCP servers for an AI-assisted Kubernetes on-call diagnosis chatbot and a Python/Flask PagerDuty automation integration. Previously at Tesla, optimized high-volume battery test data in PostgreSQL using JSONB, partitioning, and a timestamp normalization pipeline; also built PyTorch PINN training workflows and achieved a 20x speedup via batch vectorization.”

“Machine learning/NLP practitioner at J.P. Morgan who led development of a production RAG system and an entity resolution pipeline for complex financial data. Deep hands-on experience with embeddings (Sentence-BERT), vector search (FAISS/pgvector), LLM fine-tuning (LoRA/PEFT), and rigorous evaluation (human-in-the-loop + A/B testing) backed by strong MLOps on AWS (Docker/Kubernetes, MLflow, Prometheus/Datadog).”

“Software engineer at Fannie Mae (~3 years) working on high-volume loan data pipelines using AWS (SQS/S3), Java listeners, Postgres, and Python/SQL-based data quality validation. Also built a chess data collection system (leveraging experience as an International Master) with robust retry/monitoring, schema-change handling, and idempotent backfills to prevent bad data from reaching downstream systems.”

“Robotics software engineer focused on simulation-first autonomy and learning: implemented TD3 and CLIP-guided pretraining for physics-based humanoid skill learning in Isaac Gym/DeepMimic. Also built a ROS2 + dual-Docker closed-loop stack for an autonomous wheel loader in Isaac Sim, combining global planning, B-spline smoothing, and real-time NMPC control.”

“ML/Simulation engineer who productionized a Multi-Agent Reinforcement Learning system for 30+ firms at Belt and Road Big Data Company, integrating research code into an enterprise backend via Dockerized deployment and scalable data pipelines on GCP/Vertex AI. Demonstrated strong production debugging by tracing apparent network timeouts to hardware memory exhaustion caused by software state-history garbage collection issues, and built custom reward functions to model complex market dynamics (entry/exit, pricing).”

“Backend/full-stack engineer (Amazon experience) who built an AWS-based integration testing platform using Flask, ECS, Docker, and CloudWatch—cutting 1000+ test cases from ~5 hours to ~30 minutes while improving log visibility for non-engineering users. Also led a zero-downtime EU region migration with rigorous ORR testing, and built a Kinesis/Firehose/S3 + Glue/Spark replay mechanism for resilient data recovery. Side project: reproducible, cost-efficient LLM hosting platform on EKS using CDK and Karpenter for scale-to-zero.”

“Built and deployed an LLM-powered platform that turns models into scalable REST/gRPC APIs, focusing on keeping GPU-backed inference fast and stable during traffic spikes. Experienced with AWS orchestration (EKS/ECS/Step Functions), safe model rollouts, and production-grade monitoring/testing for reliable AI agents and workflows.”

“Built and deployed a production clinical support LLM assistant at Mayo Clinic using a LangChain-orchestrated RAG architecture (Llama 2/PaLM) over de-identified clinical records, integrating BigQuery with Pinecone for semantic retrieval. Focused on healthcare-critical reliability by reducing hallucinations through grounding, implementing HIPAA-aligned privacy controls (Cloud DLP, VPC Service Controls), and running structured evaluations with clinician feedback.”

“AWS AI Lab engineer who deployed a production Chain-of-Thought analytical agent for tabular reasoning, emphasizing grounded tool-constrained workflows with schema-validated intermediate outputs. Built robust evaluation/logging with step-level observability to catch regressions across model versions, and has experience scaling distributed LLM training via Slurm + DeepSpeed/FSDP with checkpointing and failure recovery.”

“Built and deployed an AI receptionist product for field-service businesses (HVAC/electrician), including real-time Jobber scheduling integrations and Twilio-based calling. Combines hands-on customer/operator shadowing with strong production engineering (queueing to handle API limits, rigorous testing/mocking, mirrored prod environment) and cross-layer troubleshooting, driving user adoption through review/override workflows.”

“Interned at Backblaze building production AI systems for incident response and security operations, including an internal LLM-powered incident triage assistant that used Snowflake + RAG over historical tickets/postmortems and delivered results via Slack and a web UI. Emphasizes reliability (PII filtering, grounding, schema validation, fallbacks) and rigorous evaluation/observability (offline replay, partial rollouts, time-to-first-action metrics, Prometheus/Grafana).”

“Backend engineer who built a new REST eligibility service at Barclays that unified siloed account logic (card/loan/deposit) and integrated with web/mobile, ultimately serving millions of users daily. Also built an end-to-end LLM-based pharmaceutical care-plan generation tool in a rapid Columbia startup competition, emphasizing configurable design, strict validation, persistence, and robust error handling.”

“Robotics software engineer who built the behavior-tree orchestrator for the Vulcan Stow robotic system, migrating from a state machine to significantly improve testability. Experienced with ROS 1 and Baidu Apollo workflows (rosbag, LiDAR/image extraction) from self-driving simulation work at LG Silicon Valley Lab, and currently focused on stable Docker/docker-compose-based deployments with disciplined QA and hotfix processes.”

“Robotics software engineer with Stanford coursework and Georgia Tech research experience, focused on end-to-end autonomy for mobile manipulation and real-time planning under uncertainty. Built a ROS 2 LoCoBot system combining Gemini speech-to-text, YOLO-based RGB-D perception, navigation, and grasping with robust synchronization/TF fixes, and developed an information-theoretic UGV planner for radiological source localization validated via Monte Carlo simulation.”

“Applied Scientist intern at Amazon who built a production-adopted LLM-judge to evaluate an agentic chatbot’s intermediate reasoning and tool calls using a knowledge-graph grounding approach. Also published award-winning work (ACM SIGSOFT Distinguished Paper) using LangChain + GPT-4 tools to generate factually grounded commit messages, with rigorous human-centered evaluation metrics.”

“ML/LLM engineer who built a production RAG system (GPT-4 + FAISS + FastAPI) to deliver fast, grounded answers from proprietary documents, optimizing for sub-200ms latency and high-concurrency scale. Strong MLOps/observability background: drift monitoring with Prometheus + Streamlit, automated retraining via Airflow, Kubernetes autoscaling, and MLflow-managed model lifecycle, plus inference cost reduction through quantization and structured pruning.”

“ML/NLP engineer focused on production-grade data and search/recommendation systems: built an end-to-end pipeline that connects unstructured customer feedback with product data using TF-IDF/BERT, Spark, and AWS (SageMaker/S3), orchestrated with Airflow and monitored for drift. Also has hands-on experience with entity resolution at scale and improving search relevance via BERT embeddings, FAISS vector search, and domain fine-tuning validated with precision@k and A/B testing.”