9 roles across AI startups, automotive, and healthcare. The teams I’ve joined, the problems we tackled, and what I built.
Party
January 2025 - Present
(formerly Maitai)
Founding Software Engineer
Redwood City, California · January 2025 - Present
A platform where humans and agents build together. Started as Maitai, making enterprise AI reliable. AI agents hit 99% accuracy
Contributions (18)
Co-led the technical pivot from Maitai to Party, reframing an enterprise AI reliability platform into a workspace where teams and agents build together, and owning the architecture calls that carried the existing runtime, evals, and versioning forward instead of starting from zero
Built the agent runtime and a structure-agnostic Bring Your Own Agent ingest, so any agent shape (OpenAI Agents SDK, Google ADK, or fully custom frameworks) can be captured, observed, and evaluated without a rewrite, cutting onboarding from days to under 15 minutes
Architected and shipped the Maitai CLI, letting developers import an existing agent in a single command through a schema-driven pipeline that auto-detects, validates, and maps agent configurations into Maitai's runtime, with async imports, reattachable import status, selective excludes, and lossless export
Designed agent versioning with immutable revisions, pinning, rollback, and safe deletion, so customer edits never mutate a live agent without rigorous testing beforehand
Unified request, workflow, and agent test sets into a single type-less test set behind one /api/v1/test-sets facade, collapsing three divergent evaluation surfaces into one and making clone and convert operations non-blocking through job rows
Built agent evals that snapshot recorded production sessions straight into test sets, turning real customer traffic into regression suites, and added post-run eval triggering plus retries so a flaky provider never invalidates a test run
Improved evaluation methods through multi-stage analysis and correction, detecting faults in model output with 94% accuracy, enhancing datasets and driving measurable improvements in downstream model performance
Diagnosed and rectified a critical evaluation bug, significantly enhancing the quality of fine-tuned models by ensuring accurate data alignment and interpretation
Shipped dataset augmentation with reasoning, embeddings, and model profiles, resulting in roughly 2.3x more training diversity and 18% higher evaluation pass rates
Developed request augmentation techniques to synthesize high-quality data, expanding semantic coverage and boosting the accuracy and generalization of fine-tuned models
Cut p99 time-to-first-token by 38% by tightening Redis stream block windows in the orchestration layer, and sped composition fetching roughly 6x through Postgres CTE rewrites
Hardened inference reliability with vLLM routing for self-hosted models, automatic provider failover when an upstream goes down, auto-disabling of models that report as missing, and a database-backed health endpoint so Kubernetes probes actually detect outages
Migrated embedding and metadata extraction onto self-hosted Qwen3-Embedding-4B and Mistral-Small instances, removing an external vendor dependency from the hot path along with its per-call cost
Designed and implemented system prompt substitution, granting users greater control over generated responses and enabling dynamic prompt engineering for improved customization and adaptability
Identified and resolved a caching bug, contributing to SOC 2 compliance by ensuring data consistency and security
Built large parts of the customer-facing Studio and portal in React and TypeScript: the agent chat interface, conversation tree editor, session timelines, assistant sidebar, dataset and test run views, Stripe checkout flow, and inline agent failure reporting
Shipped a platform-wide search endpoint with a matching CLI command, and wired auto-generated OpenAPI specs, CLI code, and docs into CI so the SDK and documentation never drift from the API
Software Engineering Intern - Applied AI / Full Stack
New York, New York · January 2025 - January 2026
Built configurable legal AI for midsize firms. The team raised a $7M Seed while I was there.
Contributions (7)
Led the development of a multi-agent system to optimize legal workflows, integrating Retrieval-Augmented Generation (RAG) and AI-driven judicial reasoning for enhanced decision-making, directly contributing to a Seed raise valued in the 95th percentile
Refactored backend architecture by introducing PostgreSQL views, reducing latency by 70% and improving Largest Contentful Paint (LCP) by 65% on average
Developed an AI-powered document review system for lawyers, leveraging Natural Language Processing and rule-based analysis to automate compliance checks against legal guidelines, processing over 1000 requests daily across hundreds of active users
Built the playbook review flow, letting a firm run its own standard positions and guidelines against an incoming document and get back clause-by-clause findings instead of reading the whole thing end to end
Built the summarize flow, turning long legal documents into structured summaries attorneys could act on directly
Implemented chunked inference to get around model context windows, splitting documents spanning millions of tokens into ordered chunks, running inference across them, and stitching the results back together with citations still pointing at the right place in the source
Leveraged NextJS and TailwindCSS to build a responsive and user-friendly interface, shipping product surfaces used by hundreds of attorneys daily at midsize firms
Built a sports analytics platform that turned athlete data into AI-driven training recommendations.
Contributions (3)
Designed and built the backend for a sports analytics platform, powering AI-driven recommendations to identify optimal drills and enhance athlete performance
Integrated Single Sign-On (SSO) using Firebase Authentication, streamlining access management for an improved user experience
Achieved a top five finish among 72 start-ups at the Michigan Ross Business + Tech Innovation Jam
Software Engineering Intern - Identity and Access Management Team
Dearborn, Michigan · January 2024 - April 2024
Developing secure authentication services and API infrastructure
Contributions (5)
Spearheaded the decoupling of critical Java endpoints enveloped in a complex monolithic application exceeding 100,000 files, leveraging microservices architecture to achieve a 60% reduction in latency and enhance system uptime
Independently implemented fixes to existing code to pass 42Crunch conformance scans and company security standards
Introduced rate limiting to mitigate API Abuse, Brute Force attacks, and DDoS attacks directed toward IAM services
Leveraged Spring Boot tools to parallelize cucumber acceptance tests and reduce deployment time by 75%, alongside CRON-scheduled automation
Maintained 100% Jacoco branch coverage and Pitest mutation test standards through accurate and thorough unit tests
Software Engineering Intern - Vehicle Services Team
Dearborn, Michigan · September 2022 - December 2022
Automating and modernizing service scheduling systems
Contributions (4)
Developed a full-stack web application using React and Node.js that automates vehicle service scheduling and inquiries, eliminating customer wait times and removing the need for manual employee intervention
Leveraged Test Driven Development through REST API validation using Jest, ensuring application robustness and reliability
Maintained continuous integration, delivery, and monitoring of pipelines using Tekton, Terraform and SonarQube
Utilised Google Cloud Platform (GCP), Firestore, and Redis for scalable data storage and real-time synchronization, reducing read latency by 41% on average and improving system reliability
Software Engineering Intern - Device Management Team
Richmond Hill, Ontario · January 2022 - May 2022
Developing cloud-based medical device management solutions
Contributions (3)
Developed a full-stack web application using Angular, TypeScript, C#, and SQL Server to efficiently manage Qvella devices and customers, deployed on Amazon Web Services (AWS) for reliable availability
Designed, tested, and validated RESTful APIs for seamless interaction with a SQL Server database for data storage and retrieval, implementing OAuth 2.0 with Microsoft Entra ID (formerly Azure Active Directory) for data security
Digitized the managing and communication of advanced Sepsis testing machines, eliminating 33% of on-site maintenance costs
Qvella
July 2020 - July 2021
Software Development Intern - Software Automation Team
Richmond Hill, Ontario · July 2020 - July 2021
Automating medical testing processes and manufacturing workflows
Contributions (3)
Directed the design and development of Python software for rapid Sepsis testing
Coordinated virtual meetings, showcases, and installations throughout the COVID-19 pandemic
Reduced pollution in labs by over 80% through the creation of paperless production lines using Python, increasing production of Positive Blood Culture samples by 70% through the automation of manufacturing
NOVX Systems
September 2019 - January 2020
Software Development Intern - Safety Solutions Team
Richmond Hill, Ontario · September 2019 - January 2020
Building workplace safety and impairment detection software
Contributions (2)
Built Angular software to administer workplace impairment tests, significantly enhancing injury detection and management
Engaged with customers to capture key requirements and translate them into measurable deliverables
