Spencer – Python, AI, AI agent development

A secure digital-legacy organizer where people store the information their families will one day need: financial accounts, legal documents, healthcare records and more. Owners can give loved ones permission-based, least-privilege access to specific parts of their legacy, and write personal letters to be shared with them. Because the platform holds highly sensitive personal data, every stored record is protected with encryption at rest.

• Designed and built the platform end-to-end in about two months, from the data model and security architecture to production deployment.
• Implemented envelope encryption across all stored records, so users’ financial, legal and health information stays encrypted at rest.
• Built the permission model for granular, least-privilege sharing of individual legacy sections with designated family members.
• Built the FastAPI backend on GCP Cloud Run with Cloud SQL (PostgreSQL) and deployed the frontend on Cloudflare Pages.
• Used AI-assisted development (Claude Code) to compress the MVP timeline.

An AI copilot for academic researchers that supports the whole grant application process except the writing itself. It recommends relevant funding opportunities based on each researcher’s profile. It runs brainstorming sessions with domain-specialized AI “researchers” and reviews draft proposals from multiple disciplinary perspectives, leaving structured feedback and in-document comments. A built-in research-report tool answers questions using only published, verified papers. The platform was sold to research universities under enterprise contracts.

• Built and shipped an AI red-team review pipeline that reviews technical grant proposals at a level comparable to human reviewers, cutting review time from 2+ hours to under 10 minutes per proposal. The pipeline covered PDF ingestion, domain-specific preprocessing, LLM orchestration and RAG-based follow-up chat.
• Secured and managed multiple five-figure enterprise contracts with top research universities. Owned client relationships end-to-end and used user interviews to shape the AI features.
• Built a hybrid RAG recommendation engine (embeddings + keyword search + reranker) that matches researcher personas to relevant grants and returns results within a few seconds.
• Architected a multi-agent brainstorming chatbot where users consult several domain-specialized AI “researchers”, each with tool and RAG access, in a single conversation.
• Migrated the backend from Flask to FastAPI within weeks to enable streaming and async. Split it into separate services for long-running AI tasks and web traffic.
• Designed a serverless + lightweight SQL architecture that kept infrastructure costs near zero while leaving a path to enterprise scale.
• Fine-tuned open-source LLMs (LLaMA, LoRA) and integrated the OpenAI, Claude and Gemini APIs. Led engineering teams of 3–6 people using agile/scrum.

A risk and compliance reporting environment for a payment-processing company. It gives the Risk department continuous oversight and audit capability over its merchant portfolio through dashboards refreshed with near-real-time data. Automated data pipelines also feed reporting for the Finance, Marketing and Sales departments.

• Built the dashboard suite the Risk department used to oversee and audit its portfolio, with near-real-time data refreshes.
• Automated data retrieval and transformation jobs for four departments using Salesforce Wave Analytics, Informatica and Pentaho.
• Migrated and optimized existing Pentaho jobs in Informatica for more efficient processing.

A research platform and data analysis effort for a major US city’s mobility innovation office. It studies how micro-mobility, such as shared scooters, interacts with public transit, to find underserved areas. The platform also helps transportation professionals explore the impact of new mobility modes, including suggesting comparable census tracts nationwide.

• Ran the statistical analysis of the city’s micro-mobility data, which led the city to expand and reallocate its scooter deployments.
• Built a cluster-based recommendation system using unsupervised learning that suggests census tracts nationwide similar to the one being viewed.
• Worked with UI/UX designers to shape the collaborative platform for transportation researchers.