New World: Aeternum

Disclaimer

Public media used.

About

Online action RPG set on the cursed island of Aeternum, featuring large-scale PvE/PvP battles, classless weapon builds, crafting-driven progression, and territory control. Shipped live content and AI systems at Amazon Games.

Link to Steam

Season of the Divide (S8) Trailer

Major Contribution: World Boss Difficulty Scaling

Nighthaven (S10) Trailer

Major Contribution: AI Communication & AI Stealth Interaction

My Works

Developed core AI navigation improvements, AI communication, and behavior systems that enhanced gameplay across multiple years.

• AI Stuck Resolution (2022): Designed and implemented a raycast-validated teleport fallback in the AI pathing pipeline to recover from physics-based stuck states without dropping characters below the world, including tunable configuration and performance telemetry to monitor impact.
• AI Communication (2023): Built a reusable behavior-tree-driven framework for AI-to-AI communication, allowing enemies to discover partners, negotiate roles, and coordinate actions via tagged queries and lightweight handshakes.
• World Boss Difficulty Scaling (2024): Architected a dynamic difficulty system that tracks participants in large open-world encounters and adjusts boss stats in real time based on player count and composition, exposed to designers via data tables for tuning.
• Stealth Combat & Decoy Threat (2025): Led design and implementation of a stealth combat feature where entering stealth adjusts AI perception and spawns decoys with rebalanced threat, including new threat-handling logic that keeps multi-enemy fights readable and fair.
• Configurable Data Table Framework (2025): Implemented a configuration framework that lets designers override game data table values via live configs, enabling rapid experimentation and balance changes without redeploying builds.
• Performance & Scale Testing Infrastructure: Created automated bot scripts, AI stress scenarios, and unit test utilities to measure AI CPU cost and behavior stability at scale, feeding results into telemetry dashboards for regression detection.

Authored 20+ Confluence pages including setup guides, runbooks, and technical briefs for features like AI Communications, Behavior Tree Best Practices, and World Encounter Scaling.

Achievement	Description	Impact
AI Pathing Resolution	Raycast teleport and navmesh validation	Fixed persistent AI pathing failures, saving 800k+ failures per month in Brimstone region
AI Communication System	Restricted messaging via behavior trees to coordinate behaviors between AIs	Enabled group tactics (e.g., boss-minion) mainly used for new region, Nighthaven
Dynamic Difficulty Scaling	Participant-based AI scaling via internal data sharing system	Balanced encounters for 1-200+ players
AI Stealth Interaction	Dynamic perception adjustment and decoy mechanics	New weapon class with refreshing gameplay
Configurable Data Table Framework	Config overrides for data table values	Eliminated deployment needs for tuning, saving time for internal playtests and hotfixes

Challenges

World Boss Difficulty Scaling

Boss fights are usually in dungeons but Open-world boss fights had to feel fair with anything from a handful of players to hundreds, but manual tuning for every population level was impossible and server performance limited how often heavy logic could run.

Approach

• Designed a participant-tracking layer that counts active contributors in an encounter and computes a difficulty scale factor from designer-authored data, rather than hard-coded formulas.
• Integrated the scale factor into AI stats via existing systems (health, damage, status effects) so scaling reused battle-tested components instead of inventing a parallel pipeline.
• Added throttling and replication safeguards so frequent scaling updates wouldn't cause excessive network traffic.

Lessons & Future Improvements

• Future work could give designers per-phase or per-creature type curves rather than equations.
• Ability for AI to self scale by reading their surroundings and applying their own stat multipliers instead of relying on encounters.

Stealth, Decoys, and Threat

A new stealth-oriented weapon, Daggers, needed to let players temporarily “disappear” or redirect attention with decoys, without making encounters trivial or breaking existing threat and aggro rules in multiplayer boss fights.

Approach

• Split the problem into three axes: perception (vision/audio adjustments while stealthed), threat adjustment (how much threat a decoy will have), and combat state (in-combat vs out-of-combat rules).
• Introduced a dedicated handler that tracks the relationship between a player and their decoy, updates threat for both in lockstep, and respects designer-tunable percentages instead of hard-coding values.
• Built test scenarios for single-player, mixed-weapon parties, and multiple stealth users to validate that aggro switches remained predictable and readable under edge cases.

Lessons & Future Improvements

• Extend the stealth config to cover more nuanced states (e.g. boss phases, environment changes) so designers can control how specific AIs perceive stealth without needing new status effects or code paths.
• Add better debugging/telemetry hooks for stealth (e.g. logging why an AI did or did not perceive a stealth target, and which table row/flag caused it) to speed up tuning and bug triage on complex encounters.

Configurable Data Table

Data table changes normally required code or data deployments, which slowed iteration on live events and made it hard to react quickly to player behavior or performance issues.

Approach

• Designed and implemented a generic framework that lets data tables read override values from configuration.
• Solved edge cases around complex cell formats (values containing commas/colons) and empty cells so designers could safely express new values without breaking parsers.

Lessons & Future Improvements

• Next steps could include a validation tool that simulates overrides in a sandbox environment and warns designers about potential conflicts before changes reach live servers.

Tooling, Automation, and Performance

Manual reproduction of complex AI and performance issues was slow and unreliable, especially on consoles. The team needed automation and tooling that could drive bots, capture telemetry, and run repeatable scenarios across platforms.

Approach

• Built generic unit test helpers to keep frame timing accurate, plus reliable repetition utilities to flush out randomized behavior.
• Created bot scripts and scenarios that spawn and drive large groups of players and AIs, then integrated them with performance capture commands to automatically produce CPU and telemetry traces on PC and consoles.
• Standardized logging and capture paths for consoles and documented easy-to-follow workflows so QA and engineers could run the same tests and retrieve the same artifacts without custom tooling per platform.

Lessons & Future Improvements

• Grow the bot-scripting toolkit into an automatic process with perf thresholds to pass/fail the scale test, so large tests surface regressions instead of relying on manual telemetry inspection.
• Have a reliable way to open console logs intuitively without needing to reference documents.

Play Demo Video