01Work
Work
AI Engineer / Full-Stack Developer at Hime, building Wheelbase: a multi-tenant dealership platform with hybrid vector search, a governed AI operations assistant, and a Go backend. Creator of Grammario and Lociros, two NLP products for language learners.
I — Experience
Where I have shipped
AI Engineer / Full-Stack Developer
Hime · Wheelbase ↗
Own full-stack delivery of Wheelbase, a multi-tenant dealership operations platform, across a React/TypeScript front end, Go and Python APIs, and Supabase Postgres, shipped to web, Electron desktop, and an Expo mobile field app from one Turborepo monorepo.
- Built production AI features: a natural-language-to-SQL operations assistant with risk-tiered write approval, vector-powered demand matching, and operational recommendations.
- Built hybrid inventory search over pgvector embeddings and Postgres full-text ranking, powering the IMX auction scoring system.
- Re-engineered backend services from Python/FastAPI to Go/Gin, cutting database load by 80%+ and bringing latency under a second with zero breaking changes.
- Designed tenant-scoped schemas and Row Level Security across 66 tables so isolation is enforced by Postgres, not application code.
- Replaced a paid VIN-decoding API with an offline Go service over a ~2GB NHTSA database, with validation and auto-correction.
- Built streaming CSV ETL for auction runlists with per-auction column mapping and 500-row batch writes, and set up GitHub Actions CI/CD with Docker and Nginx.
- Onboarded and mentored a new engineer through the monorepo, and turned requirements from finance and operations stakeholders into shipped features.
Go · TypeScript · React 19 · tRPC · Supabase · pgvector · Electron · Expo · Docker
Mar–Sep 2025
Boston, MA
AI Engineering & Data Science Intern
Massachusetts Executive Office for Administration and Finance
Built AI-assisted document workflows for state government, where output had to be accurate, traceable, and usable by non-technical staff.
- Automated multilingual translation of 200+ financial documents a month with AWS Bedrock and AWS Translate.
- Cut translation turnaround by 60% while preserving layout, deconstructing and rebuilding PDF structure with borb and PyMuPDF.
- Worked directly with cross-functional government stakeholders to deliver production-grade workflows with high accuracy requirements.
Python · AWS Bedrock · AWS Translate · PyMuPDF · borb
II — Impact
Measured, not estimated
80%+
less database load
Re-engineered Wheelbase's backend from Python/FastAPI to Go/Gin with zero breaking changes. Responses went sub-second.
~217
tRPC procedures
Across 35 domain routers, over a 66-table Postgres schema with Row Level Security on every table.
0
external API calls per VIN decode
Replaced a paid VIN API with an offline Go service over a ~2GB NHTSA database, with validation and auto-correction.
60%
faster translation turnaround
AWS Bedrock document pipelines for Massachusetts A&F, processing 200+ financial documents a month with layout-preserving PDF reconstruction.
9s → 4s
Grammario analysis latency
Parse, LLM explanation, and embedding run concurrently with asyncio.gather. The dependency tree itself renders in 300–500ms.
Merged
into pandas core
PR #64567 replaced a misleading plotting error with an accurate diagnostic, merged by a core maintainer without revisions.
III — Case studies
The long versions
- 01WheelbaseAuction intelligence, inventory, and recon for used-car dealers. Hybrid vector search ranks every auction car against real inventory gaps; a governed AI assistant runs the rest.
- 02LocirosGraded readers where A2 is actually A2. Every passage is checked by a morphological analyzer before you read it.
- 03GrammarioClick a sentence. See the structure. Universal Dependencies first, AI explanation second.
IV — Education
Study
2026–2028 (expected)
M.S. Applied Data Analytics, AI & Machine Learning concentration
Boston University
Part-time, alongside statistics and data analytics.
2019–2024
B.S. Computer Science, Data Science concentration
University of Massachusetts Lowell
Operating Systems, Databases, Computer Architecture, Artificial Intelligence, Natural Language Processing, Analysis of Algorithms, Data Structures.
V — Certificates
Coursework
Neural Networks and Deep Learning
DeepLearning.AI
Natural Language Processing
DeepLearning.AI
Programming with Google Go
University of California, Irvine