neueHCT internship
I built AI coding workflows and internal platforms at an autonomous-driving company, with nine projects deployed.
- Role
- Software engineering intern, AI Infrastructure
- Team
- neueHCT, Shanghai (a Continental and Horizon Robotics joint venture)
- When
- May – Aug 2026
- Outcome
- 11 projects, 9 deployed
- Built with
- Python, FastAPI, Vue.js, TypeScript, Claude Code, C++
LLM feature requests
Architected the core workflow of a deployed AI feature-request platform.
I designed generation, independent AI review and running previews around one branch per request, giving developers control over the merge. I also built revision, cancellation and recovery across Windows and Linux.
More detail
A colleague continued development and handled the production release. My implementation also preserved the previous completed draft when a revision failed or was cancelled.
C++ test generation
Fixed and extended the shared LLM pipeline for autonomous-driving C++ tests.
I removed compilation blockers, corrected test-target detection and repaired coverage-reporting races. I also built compiler-error recovery and a persistent batch supervisor, so generation could repair failures and continue unattended through reboots.
More detail
- Replaced class-name guessing with tree-sitter parsing of included headers, clearing repeated compilation failures.
- Fixed exception handling and conflicting entry points so generated Google Test suites could compile and run.
- Owned planning-and-control and perception generation in a five-person team. More than a thousand results from my runs were reused in the team’s full-codebase generation, and my code changes remained in the shared pipeline.
Hardware verification
Replaced the team’s spreadsheets with its official hardware-verification system.
I took it from a basic tracking request to a shared system for inspections, photos and part histories. I defined the workflow with users, built the software and led migration and training; non-developer colleagues made hundreds of updates.
More detail
I added conflict checks so inspectors could work on the same record without silently overwriting one another’s changes.
Workplace AI and tooling
I built and released a workplace AI assistant, a production platform for managing employee access to AI tools, and a self-service survey system adopted by the AI team.
View contributions
- Built the assistant’s custom tool loop so the model could choose and chain workplace API calls, retrieving calendar, task and conversation data with each employee’s own read-only permissions.
- Built a self-service survey platform and led its adoption within the AI team. A company-wide campaign received replies from about four in five employees contacted.
- Designed single-use access handoffs tied to the assigned employee, with unused links revoked on reassignment and usage history preserved.
- Delivered bilingual task workflows, task suggestions and onboarding; authored an internal research report on subquadratic attention.