Nick Lemoff

Founding Engineer

• Architected Evie, FullProof's real-time deposition agent, from a demo spread across disconnected services into a unified GCP-backed production system for live transcript processing, document-grounded reasoning, attorney-facing suggestions, and deposition-prep workflows.
• Owned FullProof's applied AI research loop: papers, open-source systems, model releases, and hundreds of prototypes across agent memory, real-time agents, hybrid RAG, retrieval/reranking, tool use, long-context workflows, and eval methodology.
• Productionized research ideas into a Python/FastAPI agent service with streaming endpoints for live transcript lines, chat/RAG, document extraction, strategy generation, outline generation, tracker summaries, persistent case memory, and final-round deposition workflows.
• Designed real-time agent orchestration that records transcript state, routes each line through research / quick-check / suppression paths, and returns citation-backed tactical suggestions with timing, route, usage, and tool-call traces.
• Built document-grounded retrieval over case files with Vertex AI Search / Discovery Engine, case-scoped filtering, dynamic exhibit mapping, caching, concurrency controls, source-identity observability, and fail-closed behavior for unsupported evidence.
• Created replay/evaluation infrastructure for AI reliability: transcript replays, browser-projected agent output, service health snapshots, event logs, line-level lifecycle traces, scoring, and regression coverage.
• Shipped across React/Vite, TypeScript, Node/Express, Python, FastAPI, PostgreSQL, Prisma, Redis, Socket.IO, BullMQ, Vertex AI Search, and Google Cloud.

Simulations Team Lead

• Led a 25+ person simulations team across 7 vehicle-modeling projects for UC Berkeley's Formula SAE electric racecar team.
• Developed a lap-time simulator using a two-track bicycle model and introduced Git/GitHub workflows for a larger engineering contributor base.
• Built aerodynamic and vehicle-analysis pipelines that connected CAN data with simulation outputs.
• Containerized simulation environments with Docker so contributors could work from a consistent setup.

Research Intern

• Implemented optimized graph edit distance code in Python and C++ for inorganic materials datasets after reviewing core algorithmic approaches.
• Contributed to a long-running materials-science project embedding 167,500 ICSD entries into a structural-similarity space to study experimental discovery, computed materials proposals, and synthesizability priors.
• Reviewed and merged multiple PRs to make the latest research code reproducible and runnable on other machines ahead of the group's Nature submission.

Machine Learning Student Assistant

• Developed Faster R-CNN and YOLOv8 models to detect oil and gas wells in historical maps, reaching 0.98 mAP in under 5 minutes of training.
• Reviewed and compared 20+ CNN architectures for historical-map object detection.
• Co-authored an abstract submitted to AGU 2024 on the object-detection approach.

Data Science and Data Management Intern

• Designed Python/MySQL data pipelines for quantum dot life tests that became the company's standard processing workflow.
• Built Tableau dashboards for quantum dot validation data and reduced manual review work.
• Wrote Python and SQLAlchemy monitoring code to detect and report issues in company MySQL databases.

Simulations Engineer / Data Scientist

• Developed a tire model using Tire Testing Consortium data, pandas, and scikit-learn to guide next-year vehicle parameters.
• Created a battery-pack sizing process that balanced sizing constraints, weight, and lap-time impact.