Vetted Data Scientists

“Backend engineer who built and evolved Python/FastAPI services (including AWS-deployed ML prediction APIs) for real-time profitability and risk insights at TenXengage. Emphasizes pragmatic architecture, strong validation/observability, and secure access controls (RBAC + row-level filtering), and has led safe migrations via parallel runs and incremental rollouts; reports ~20% forecasting accuracy improvement.”

“Built a LangGraph-based, tool-routing LLM chatbot to deliver fast, trustworthy investment-stock insights (including tariff impact) and deployed it to production on Snowflake after initially developing in Azure with AI Search and the Microsoft Agent Framework. Improved routing robustness by moving from LLM-based decisions to a deterministic router backed by schema-relationship graphs and YAML metadata, and ran the project iteratively with non-technical stakeholders over an 8-month engagement.”

“Built a production-grade LLM evaluation and regression system that stress-tests models across hundreds of iterations, combining LLM-as-judge, semantic similarity, statistical metrics, and rule-based checks, with results delivered via stakeholder-friendly HTML reports and dashboards. Experienced orchestrating multi-agent RAG workflows using LangChain/LangGraph and event-driven GenAI pipelines in n8n integrating OCR, speech-to-text, and external APIs, with strong emphasis on reliability, observability, and explainable failures.”

“Built a production-grade, semi-automated document recognition and classification system for large volumes of scanned PDFs, starting from little/no labeled data and handling highly variable scan quality. Deployed on AWS using SageMaker + Docker and orchestrated on EKS with a microservices design that scales CPU-heavy OCR separately from GPU inference, with strong reliability controls (validation, fallbacks, retries, readiness probes).”

“Data scientist/researcher who has built two multimodal LLM systems: an AI-assisted medical triage pipeline using GPT-4o vision + RAG with confidence-scored red/yellow/green outputs, and a master’s project on multimodal cyberthreat detection combining multiple models and using TinyLlama to generate human-readable risk reports. Also partnered with business analysts at Sanvar Technologies to deliver a churn prediction pipeline and Tableau dashboard for decision-making.”

“LLM engineer/data analyst who built a production RAG QA assistant over the Jurafsky & Martin NLP textbook to reduce hallucinations and provide explainable, source-grounded answers. Experienced with LangChain/LangGraph orchestration, retrieval optimization (embeddings, vector DBs, caching), and rigorous evaluation/monitoring (Retrieval@K, A/B tests, telemetry/drift). Previously communicated analytics insights to non-technical stakeholders at GS Analytics using Power BI and simplified reporting.”

“Internship at Discovery Education building a production LLM/RAG chatbot that let marketing and sales teams query and interpret Looker/BI dashboards in natural language, with responses grounded in compliance and state education standards. Emphasizes rigorous evaluation (faithfulness/precision/recall/latency) plus user-feedback analytics, and used LangChain for orchestration, chunking/context-window control, and integration with enterprise sources like SharePoint.”

“Data scientist who has built production LLM agents (GPT-4o + LangChain + RAG) to automate analyst-style ad hoc CSV analysis with guardrails and GPT-as-a-judge evaluation. Also delivered an explainable healthcare NLP system for ICD code classification by collaborating closely with clinicians, using a hybrid rule-based decision tree + BERT model to reach 97% accuracy and cut manual review time.”

“Built and deployed a production generative-AI copilot at Tungsten that automates invoice/form extraction template creation, reducing weeks of manual model-building work. Combines fine-tuned LLMs (PyTorch/HuggingFace) with OpenCV layout grounding to reduce hallucinations, and runs an end-to-end Kubeflow-based MLOps pipeline with drift monitoring, canary releases, and automated retraining.”

“Robotics engineer focused on simulation, modeling, and control, with hands-on sim-to-real experience from a soft, foldable “grasshopper” robot where friction/contact physics and servo dynamics drove real-world performance gaps. Built a ROS 2 voice-operated TurtleBot system integrating YOLOv5 + stereo depth for object picking with an attached arm, and debugged AMCL/SLAM to cut localization error from 10–13 cm to ~5 cm. Currently developing a quadruped in MuJoCo with a 3-layer control stack (RL + MPC + PD) and an RL training pipeline in JAX ahead of hardware.”

“Built and deployed production LLM applications including a natural-language-to-read-only-SQL system focused on ambiguity handling and query safety (schema whitelisting, intent validation, confidence checks, deterministic execution). Experienced with LangChain-based, modular agent orchestration and RAG document QA for large PDFs, with a metrics-driven testing/evaluation approach and cross-functional delivery with marketing on an AI content recommendation/search tool.”