Software Engineer — Full-Stack & Game Dev
I build robust systems across multiple domains: from game development with Unity/Unreal to machine learning models and data structures. Focused on performance, scalability, and clean architecture.
Projects
Polished slices with proof: demos, metrics, and short write-ups.
Dhammapada — Unity FPS Dungeon Game
Gameplay/AI/Systems (Team Lead)
Immersive first-person dungeon crawler with custom enemies, traps, weapons, and boss encounters. Built end-to-end by a 3-person team with zero external assets—all 3D models created in Blender from scratch.
- Led team of 3; coordinated gameplay, AI, and asset pipeline with Blender integration
- Enemy AI with patrol, chase, and attack states; dynamic combat encounters with melee/ranged variety
- Interactive dungeon elements: trap doors, hidden passages, environmental hazards, player feedback
- Player systems: health, damage, melee + ranged combat, weapon pickups, inventory management
- C# core mechanics, collision handling, door triggers, level flow, and state management
- Boss fight with unique attack patterns, phases, and boss-specific AI routines
- 100% custom 3D assets: enemies, weapons, environment props, and decorative elements modeled in Blender
Crown, Fire & Conquest — In-Progress Strategy Game
Real-time strategy game inspired by Age of Empires and Age of Mythology. Features multi-unit selection, RTS-style controls, dynamic combat, enemy AI, and a building placement system with visual feedback.
- Multi-unit selection with drag-box and click mechanics for squad management
- RTS-style right-click movement using Unity NavMesh for fluid unit pathfinding
- Basic combat system with health, damage, and unit death mechanics
- Simple enemy AI behavior to create dynamic skirmishes and combat encounters
- Building placement system with transparent 'ghost' preview and valid/invalid placement feedback
- ScriptableObject data architecture for scalability; profiling-focused optimization
- In-progress economy and tech tree systems; growing toward full production-quality RTS
Unreal Engine Action-Adventure (3rd Person)
Narrative-driven desert world with exploration, combat, UI flows, day–night cycle, and populated NPC behaviors.
- Hybrid Blueprint + C++ core mechanics; reusable components & subsystems
- Full UI menus (start, settings, inventory, level-up) with controller support
- World building: oases, settlements, authored foliage; cinematic lighting & post-processing
- Dynamic environment: day–night, ambient SFX, NPC schedules
- AI & NPCs: patrol, aggro, crowd reactions; melee, block, bow & arrow; stamina management
Memory Dive — Cross-Platform VR & PC Co‑Op Mystery
Server-authoritative multiplayer (Photon Fusion) with asymmetric PC+VR roles, XR Interaction Toolkit, and puzzle-driven environments.
- INetworkRunnerCallbacks, NetworkTransform, and custom prefabs for smooth sync
- Low-latency architecture; minimized desync; platform-tailored interactions
- Co-op systems: shared clues, event triggers, custom UI to encourage communication
- Optimized scenes for VR + PC performance; real-time networking instrumentation
B-Tree Implementation
Implemented and optimized a B-tree data structure (order m ≥ 3) in Python with high-performance operations and robust balancing.
- Developed efficient rotation, split, and merge strategies to maintain balance and optimize performance with overfull/underfull nodes.
- Engineered logic to place keys accurately within nodes, applying sibling rotations before splitting to minimize unnecessary restructuring.
- Designed merging and rotation procedures with sibling nodes to correct underfull conditions and preserve tree balance after removals.
- Created a clear path-tracing mechanism to visualize how lookups traverse the tree, simplifying debugging and verification.
- Conducted extensive local testing to confirm adherence to B-tree invariants, strengthening skills in algorithmic reasoning.
🧠 AI-Driven Spinal Condition Classification
Developed an end-to-end KNN-based machine learning model to classify spinal conditions from six biomechanical attributes.
- Conducted thorough data cleaning, outlier detection via IQR, min-max scaling of predictors, and target balancing with SMOTE.
- Performed univariate and bivariate analysis to understand feature distributions and class relationships.
- Tuned K values to find optimal parameters, achieving balanced train/test errors and ~94% accuracy at K=11 or 13.
- Documented variable definitions to improve interpretability for healthcare practitioners and researchers.
Professional Experience
Ownership, innovation, and shipping features end-to-end.
InZone — Software Engineer & Designer
- Collaborated on building InZone, a personalized AI social app (7,500+ downloads); implemented cross-platform Flutter UI/UX supporting thousands of concurrent sessions.
- Built and optimized a 3D avatar asset pipeline (glTF/GLB) with mobile performance tuning and engineered scalable feed experiences combining Unity real-time rendering with pre-rendered loops.
- Integrated GPT-based AI chatbots and prompt engineering to boost engagement time by 30%+, and implemented Firestore with advanced queries for real-time data sync across 7,500+ users.
Department of Computer Science, University of Maryland — Teaching Assistant
- TA for CMSC351 (Algorithms) and CMSC425 (Game Programming): assessed and graded assignments/exams for 300+ students, providing constructive feedback that improved student engagement.
- Held 50+ office hours covering algorithm analysis, graph/sorting techniques (CMSC351) and Unity gameplay systems & debugging (CMSC425).
University of Maryland — Research Assistant
- Fine-tuned Qwen2.5-1.5B with LoRA and built a custom RAG pipeline for CFA exam, BloombergGPT, and XBRL tasks to develop high-accuracy financial analytics agents.
- Improved model accuracy by 18–22% on CFA Level I question sets (from ~63% to ~77%) and achieved F1 scores above 85% on Financial Phrasebank sentiment classification.
Ernst & Young — Data Science Intern
- Worked on the Strategy & Transaction team to evaluate clients’ strategies and capital agendas; implemented an advanced AI model (logistic regression) achieving 93% accuracy.
- Analyzed customer attributes (spending behavior, loan status) to help marketing target high-conversion prospects and assisted the bank in increasing borrower base via targeted loan interest strategies (+56%).
OMSE, University of Maryland — Tutor
- Delivered tutoring sessions for MATH120/140/141, PHYS121, CMSC131/132 and supported 150+ student hours to improve understanding and performance.
More About Me
I am currently in my senior year of college. I m a very outdoorsy person , When I'm not building projects, I am probably hiking, climbing or playing soccer. Currently I m playing field hockey in college which is really fun. Ofcourse, I also love gaming growing up I used to play strategic game like Rise of Nation(let me know If you have played this I have never found a single person who has played this), Age of Empires. Now I mostly play FPS games like CS and Valorant. Also if you are a game of thrones fan the url has the game of thrones reference :D. I love anime and other animated series like arcane.
Contact
Open to gameplay/tools roles; happy to share a short reel and builds.
© 2025 Ishaan Gupta — Built with React & Tailwind.