02 / systems
What I've built
Systems built with real constraints, real metrics, and real tradeoffs.
featured systems
SYS-001 · v1.2
production ai/ml
Multimodal Inference Pipeline
// realtime edge ai preprocessing · The Plug Co
Realtime image and audio preprocessing pipelines for AI inference at the edge. Structured logging, benchmarking, and iterative optimization across the preprocessing pipeline until p99 preprocessing latency hit 11ms.
[camera/audio_input] → [preprocess()] → [normalize()] → [model.infer()]
↳ [structured_log] + [latency_tracker] + [validation_layer]
11ms
p99 latency
TensorFlow
framework
edge
deployment
SYS-002
production automation
Document Intelligence RAG
// ocr pipeline + retrieval-augmented generation · Outamation
OCR-based data extraction pipeline feeding into a LlamaIndex RAG system. Handles 1,000+ documents and 10,000+ data points. Reduced manual review workload by ~60% during validation workflows, improving data quality.
[pdf/scans] → [ocr_extract()] → [LlamaIndex] → [query_api]
↳ [data_validation] + [quality_checks]
60%
less review
10k+
data points
1k+
documents
SYS-003
project observability
Predictive Model Monitor
// ml observability + drift detection · personal project
Model monitoring system that tracks data drift, prediction stability, and inference latency in realtime. Designed around observability principles — structured metrics, persistent storage, and a live dashboard for ML pipeline visibility.
[model_output] → [drift_detector] + [latency_tracker]
↳ [FastAPI] → [SQL_store] → [Streamlit_dashboard]
FastAPI
api layer
realtime
monitoring
SQL
persistence
SYS-004 · v2.0
project async orchestration
TrustChain AI Validation
// ai-powered financial document validation · personal project
Async receipt validation system with multi-model AI backend. Node.js API layer, job queue orchestration, MongoDB persistence, Google Gemini + TensorFlow validation pipeline. ~85% validation agreement across receipt samples. Full architecture case study available.
[React_UI] → [Node.js_API] → [Gemini + TF]
↳ [validation_pipeline] → [MongoDB]
85%
accuracy
Gemini
llm
async
architecture
Deep-dive: TrustChain AI — Architecture Case Study Full system design, validation pipeline, request flow diagrams, and engineering decisions.
Read case study →
experience log
May 2026
present
Kyndryl
python developer intern · aws · backend automation · toronto
Backend automation, AWS cloud migration, and scalable system integration supporting reliable production data workflows and performance optimization.
AWS · backend automation · cloud migration
Mar 2026
present
The Plug Co
ai & immersive systems engineer · tensorflow · pytorch · remote
Realtime multimodal preprocessing pipelines for image and audio AI inference. Benchmarked and validated until p99 preprocessing latency hit 11ms at the edge.
11ms p99 image preprocessing latency
Sep 2024
Apr 2026
Toronto Metropolitan University
program admin & operations lead · python · sql · apis · toronto
Built Python automation using Google APIs and a WhatsApp bot, improving data accuracy by 95%, saving 12+ hours/month, and enabling reliable data-driven decision-making.
95% accuracy · 12+ hrs/month saved
Jun 2025
Aug 2025
Kavodax
software intern · python · apis · docker · linux · oshawa
Built backend integrations and transaction processing workflows for blockchain and banking/remittance APIs. Reduced onboarding time by 25% through systematic testing, debugging, and robust error handling.
25% onboarding time reduction
May 2025
Jul 2025
Outamation
automation extern · python · sql · data processing · remote
OCR-based data pipelines and RAG system (LlamaIndex) processing 1,000+ documents and 10,000+ data points. Improved data quality and reduced manual review by 60%.
60% reduction in manual review
Mar 2025
May 2025
BigPod Limited
software engineer · backend & systems integration · london, on
Integrated Node.js and FastAPI backends with asynchronous operations and caching, improving system performance, scalability, and maintainability for production-ready applications.
async · caching · production backend