Software Engineer & ML Enthusiast | Product Manager
HarshilChudasama
Bridging distributed infrastructure and high-performance machine learning. Building systems that are resilient at scale and optimized for latency.

MSCS (AI/ML)
Northeastern
About
Like smoke dissolving into air,the best systems are felt but not seen.
I don't just write code; I design systems with specific constraints. Distributed consistency through Saga patterns and optimistic concurrency control. High-performance ML with quantization and shared-memory IPC to hit sub-50ms deadlines.
My work lives at the intersection of Backend Engineering, ML Infrastructure, and Quantitative Systems. Designing for failure with circuit breakers, backpressure, and idempotency. Systems that adapt, flow, and simply work.
Core Expertise
Capabilities
Technical Arsenal
Tools I wield to build resilient, high-performance systems.
Languages
Backend & Systems
Infrastructure
AI & Math
Engineering Focus
Distributed Consistency
Saga, OCC
High-Performance ML
INT8, ONNX
Algorithmic Systems
Bandits, MC
Reliability
Circuit Breakers
Selected Work
Projects
Systems built with intention. Each project explores the balance between performance and elegance.

Mobile
|Mobile Engineer
QuickVerify SDK
Cross-platform React Native SDK for biometric authentication and document capture. Native modules (Swift/Kotlin) expose Face ID/Touch ID and camera edge-detection to TypeScript, with NativeEventEmitter callbacks for real-time processing feedback.
ML & AI / Distributed Systems
|ML Systems Engineer
Anytime Inference
Deadline-aware inference router that selects model quantization levels (FP32/INT8) based on real-time CPU pressure, maintaining sub-50ms p99 latency for 500+ concurrent users via stress testing.
Quantitative / ML & AI
|Quant Dev & Systems Architect
Execution Copilot
Low-latency algorithmic trading engine using lock-free ring buffers to minimize tail latency. Optimized order routing using Contextual Bandits (LinUCB), validated via walk-forward backtests on historical market data.

ML & AI / Mobile
|Full Stack Engineer
FaceFit AR
Browser-based AR eyewear try-on. MediaPipe Face Mesh feeds dense landmarks into a linear algebra pose solver (orthonormal basis from eye/forehead vectors) and a Three.js pipeline that scales/positions GLB frames at 60 FPS. Containerized frontend+API for reproducible demos.
Experience
Where I have contributed
Sep 2025 - Present
Vancouver, Canada
Software Engineer Intern
ICBC - Insurance Corporation of British Columbia
- Refactored legacy monolithic modules into stateless microservices on AWS, implementing horizontal pod autoscaling (HPA) to handle traffic bursts of 5k+ RPS
- Optimized API Gateway throughput via FastAPI and connection pooling, reducing request overhead latency by 40ms per call
- Designed type-safe Data Access Layer using generic repositories, reducing code duplication by 40%
Mar 2025 - Jun 2025
Vancouver, Canada
AI Research Assistant
Northeastern University - Mixed Reality Lab
- Engineered low-latency multimodal pipeline combining YOLOv8 with LLM reasoning, achieving ~200ms end-to-end latency
- Reduced edge model memory footprint by ~60% via INT8 post-training quantization and execution graph pruning
- Built reproducible training pipelines (PyTorch, ONNX Runtime) with versioned configs and metrics logging
Sep 2024 - Dec 2024
Vancouver, Canada
Software Engineer
BC Cancer Foundation x Northeastern University
- Implemented probabilistic record linkage engine using Bloom Filters and weighted Levenshtein distance, reaching 98.2% precision on 500k records
- Designed hierarchical RBAC system with cached permission lookups in Redis, keeping auth checks under 5ms
- Introduced structured audit logging for compliance traceability
May 2024 - Aug 2024
Boston, MA
Software Engineer Intern
Thermo Fisher Scientific
- Diagnosed and resolved thread starvation in high-throughput Spring Boot services, improving throughput by ~30%
- Optimized PostgreSQL with composite indexes and Write-Through Redis cache to offload 80% of read traffic
- Authored design docs and runbooks creating repeatable playbook for latency regressions
Recognition
Awards
Microsoft x Qualcomm On-Device AI Hackathon
Built a Snapdragon-powered navigation assistant that fused YOLOv8 vision with on-device speech guidance, offloading inference to DSP/NPU accelerators to achieve ~40ms latency.
IMC Prosperity 2 Trading Competition
Designed a market-making strategy using Monte Carlo simulations and risk-aware inventory management, outperforming baseline agents in a volatile simulated exchange.

