Vetted Graduate Research Assistants

“Insurtech R&D leader turned CTO with 25+ years across telecom, DoD, automotive, and commercial insurance. Built patented AI-driven insurance document ingestion (>95% accuracy with 1–3 samples) and led Azure-based backend/service development while managing up to 15 engineers at a commercial P&C MGA that reached a ~$60M book of business (~30 loss ratio) before run-off due to paper issues. Known for hands-on discovery (including field inspections) and rapid MVP delivery, including a risk-inspection iPad app that auto-generated draft reports.”

“Backend engineer with experience at Razorpay and Groww, focused on hardening high-throughput financial systems for reliability and low tail latency through incremental improvements (SQL/index tuning, Redis caching, timeouts, idempotency). Also built/refactored a commodity risk tracker using Supabase Auth + Postgres RLS for strict per-user isolation, with a strong emphasis on API contracts, observability, and safe migrations.”

“Backend engineer who built an AI voice-driven calendar assistant using Python/Flask with a multi-agent architecture (CrewAI + OpenAI), including agent-specific memory management and summarization to handle token limits. Also has production performance optimization experience at Trendyol, adding Redis caching to speed up an advertisement retrieval endpoint and improve page load experience.”

“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 a production multimodal LLM system (Gemini on GCP) to automate behavioral coding of family-involved science experiment videos, including preprocessing for inconsistent lighting/audio and LangGraph-orchestrated parallel workflows. Also developed rubric-based AI grading workflows and partnered closely with non-technical education stakeholders through explainability-focused walkthroughs and manual-vs-AI evaluation alignment.”

“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.”

“Built and deployed an LLM-powered claims-document summarization system (insurance domain) that cut agent review time from 4–5 minutes to under 2 minutes and saved 1,200+ hours per quarter. Hands-on across orchestration and production infrastructure (Airflow retraining DAGs, Kubernetes, SageMaker endpoints, FastAPI) and recent RAG workflows using n8n + Pinecone, with a strong focus on reliability, cost, and explainability for non-technical stakeholders.”

“Built and deployed an LLM-powered financial/regulatory document analysis platform at State Street, combining fine-tuned transformer models with a RAG pipeline over internal knowledge bases. Owned the productionization stack (FastAPI, Docker, SageMaker, Terraform, CI/CD) plus monitoring for drift/latency/hallucinations, delivering ~40% faster analyst review and improved reliability through chunking/embeddings and grounding.”