Education
M.S. Computer Science (AI/ML)
Northeastern University
About
Harshil Chudasama
Software engineer building ML systems, ranking systems, and high-throughput backend services.
I work across recommendation and search systems, multimodal inference, Java and Spring Boot services, and performance-critical Python and C++ paths. On the ML side, quantization, ONNX serving, and the queueing math that keeps p99 honest under load.
Before engineering I worked in product management and business analysis. That's why I think in user outcomes, latency budgets, and operational tradeoffs, not just system diagrams.
Currently focused on
ranking systems
Northeastern · MS CS
Hover · 5 nearest points · illustrative of the retrieval that runs my CineMatch case study
What drives me
Products where ML reaches users at scale.
Outside of work
Training at the gym
Building a jewelry brand
Reading about markets
Exploring new cities
Tennis
ML Systems
PyTorchONNXQuantizationLangChainLangGraphpgvector HNSWElasticsearch kNN
Backend & Distributed Systems
JavaSpring BootFastAPIgRPCKafkaPostgreSQLRedis
Performance Engineering
C++17/20SIMD (AVX2)Lock-freepybind11JFR / async-profilergdb / perf
Data & Analytics
PythonNumPy / pandasSQLTime-seriesMonte CarloHypothesis testing
Projects
Selected Work
Eight case studies across ranking, retrieval, low-latency inference, and distributed backends. Click any tile for the architecture write-up and the measured results.
08 projects
01/ 08
FeaturedML & AIPersonalized Recommendation System
CineMatch
Two-stage retrieve-then-rank engine. pgvector HNSW retrieval over 1536-dim taste embeddings feeds a LambdaMART re-ranker at ~0.9 ms p95. NDCG@10 of 0.814 on offline eval, +14% over popularity. Live at cinematch.harshilc.com.
Next.jsGoPythonFastAPISupabase pgvectorLightGBM
02/ 08
QuantAlgorithmic Trading Execution Engine
Execution Copilot
Simulated execution engine comparing classical, bandit, and RL policies against a persistent FIFO limit-order book. PPO and POV lead on implementation shortfall (net of fees). Optional C++ matching engine over a lock-free SPSC ring buffer.
PythonC++FastAPI
03/ 08
ML & AIAdaptive, Deadline-Aware Model Serving
Anytime Inference
Latency-bounded serving with adaptive precision selection. Python control plane + C++ ONNX runtime; an M/M/1 admission controller routes each request to FP32 or INT8 based on live CPU pressure. Benchmark shows ~99% completion where FP32-only falls to 3%.
PythonC++ONNX Runtime
04/ 08
Distributed SystemsEvent-Driven E-Commerce Backend
Commerce Services
SAGA-orchestrated checkout across inventory, orders, and payments. Optimistic locking cut observed deadlocks by around 40% under fault injection. Outbox-pattern events published to Kafka.
JavaSpring BootKafka
BC
05/ 08
Full StackDonor Engagement Platform
BC Cancer Foundation CRM
Donor management CRM for the BC Cancer Foundation. Weighted multi-criteria matching across interests, geography, donation history, recency, and capacity, with explainable breakdowns. Full RBAC and immutable audit trail.
TypeScriptNestJSReact
06/ 08
Search & RankingHybrid BM25 + Vector Search
Semantic Search Service
Spring Boot service blending BM25 and dense vector retrieval into a single relevance score, tunable per surface via scoring profiles. Ships with an evaluation harness for MRR, NDCG@k, and Recall@5.
JavaSpring BootElasticsearch
07/ 08
ML & AIBrowser-Based Virtual Try-On
FaceFit AR
Real-time eyewear try-on in the browser. MediaPipe Face Mesh (468 landmarks) into a Three.js overlay, with pose computed by classical linear algebra instead of a PnP solver. Sustained 60 FPS on a mid-range laptop.
JavaScriptThree.jsMediaPipe
2
08/ 08
MobileMobile Identity Verification
QuickVerify SDK
React Native SDK wrapping biometric authentication (Face ID, Touch ID, BiometricPrompt) and document capture (AVFoundation + Vision, or MediaStore) behind a single TypeScript API.
React NativeSwiftKotlin
Experience
Work history
Career arc · 2017 to now
EngineeringResearchProduct & Ops
ICBC
Northeastern MRL
Northeastern TA
Thermo Fisher (SE)
Felix Payments
USF Sales Ops
Thermo Fisher (BA)
St. Joseph's Research
20172019202120232025Now
EngineeringResearchProduct & Ops
SEP 2025 - PRESENT
Vancouver, BC
Software Developer Intern
Insurance Corporation of British Columbia (ICBC)
MAR 2025 - JUN 2025
Vancouver, BC
AI Research Assistant
Northeastern University, Mixed Reality Lab
SEP 2024 - DEC 2024
Vancouver, BC
Graduate Teaching Assistant
Northeastern University
MAY 2024 - AUG 2024
Boston, MA
Software Engineer Intern
Thermo Fisher Scientific
JAN 2023 - AUG 2023
Vancouver, BC · Hybrid
Product Manager
Felix Payment Systems
Led A/B testing and product roadmap for payment products across Canadian and U.S. merchants. Worked across engineering, design, and ops to ship features end-to-end.
SEP 2021 - DEC 2021
San Francisco Bay Area
Internship
Sales Operations Analyst
University of San Francisco, School of Management
Supported product launch operations and sales pipeline analysis for the School of Management.
MAY 2019 - DEC 2021
Remote
Full-time
Business Analyst
Thermo Fisher Scientific
Automated business reporting and operations workflows across multiple teams, eliminating manual handoffs and improving turnaround.
MAY 2017 - APR 2019
Hamilton, ON
Full-time
Researcher
St. Joseph's Healthcare Hamilton
Researched 4-PBA as a pre-clinical drug for adult hypertension. Analyzed clinical data with Excel, SQL, and Python under ethical research practices.
Awards
Honors & recognition
AwardAPR 2025
Microsoft x Qualcomm On-Device AI Hackathon
Microsoft x Qualcomm
1st Place
Real-time on-device AI navigation system for wheelchair users on Snapdragon X Elite. Fused YOLOv8 and Whisper with sub-40 ms end-to-end latency through DSP and NPU offloading.
YOLOv8WhisperSnapdragon X EliteDSP/NPU
AwardNOV 2024
Data for Good Hackathon
Telus × City of Toronto × Northeastern University
Runner-up
Built ParkSmart, a real-time urban parking optimisation PoC using Telus mobility data for lot availability predictions. Interactive heatmaps and routing algorithms applied to high-density downtown Toronto traffic.
Telus Mobility DataPythonNumPyPandasNext.js
Read contextAwardNOV 2024
Climate Resiliency Hack: Lightning Strikes
Northeastern University
3rd Place
AI-driven integration platform to help indigenous communities anticipate, prevent, and mitigate the effects of severe weather events.
Next.jsOpenAI APIFigma
Read contextAwardAUG 2024
IMC Prosperity 2 Trading Competition
IMC Trading
Rank 87 / 13,500 (Top 1% Global)
Python-based market-making and statistical arbitrage strategies. Linear regression for fair-value estimation, mean-reversion signals on residuals, and inventory-aware quoting logic that adjusted spreads based on position risk.
PythonNumPypandas
Contact
Get in touch
Best reached by email. I read everything, though replies may take a day or two.
Currently
Interested in AI-powered products, personalization, and discovery.
Colophon · Built with Next.js 16 · Cormorant Garamond + JetBrains Mono · d3-force · Canvas 2D · Vercel · Supabase pgvector for CineMatch. All figures cited from the linked project sources. Press ⌘K to search this site.