# K-M Samiul Haque

> Staff Software Engineer · AI/ML Systems · Platform Engineering

Staff Software Engineer at RBC in Toronto. At the bank since 2017, working across developer tooling, SRE and chaos engineering, security platforms, and now AI. I build LLM ensemble labelling pipelines, fine-tune embedding models, and run intent classification on CPU-only infrastructure.

## Links

- Website: https://www.samiulhaque.com
- LinkedIn: https://www.linkedin.com/in/samiul-haque
- GitHub: https://github.com/samhaque
- Email: sammy.haque@alumni.utoronto.ca
- Resume: https://www.samiulhaque.com/static/content/K-M_Samiul_Haque_Resume_2025.pdf

## About

At RBC since 2017. I started as an intern and I am now a Staff Software Engineer.

I joined RBC in 2017 as an intern, straight out of a Statistics degree, wanting to see how software actually runs in production. I stayed. Since then I have worked on developer tooling and ChatOps, site reliability and chaos engineering, security platforms, and most recently AI systems.

Right now I own the intent detection and classification models for the bank's call-centre platform: labelling, training, and inference. The work covers LLM ensemble labelling pipelines, knowledge distillation for embedding models, and a real-time classifier that labels calls as they progress, served on CPU-only infrastructure. Most of the effort goes into making research techniques hold up under real production constraints.

My background is a Statistics degree from the University of Toronto, so I tend to reach for the data before the opinion. I want to understand why a model behaves the way it does, not just whether it clears a benchmark.

Outside work I run a homelab, tinker with networking, and build small things to test ideas.

## Current Role

**Staff Software Engineer** · AI Advice Centre, Royal Bank of Canada (Sep 2025 to Present)

- Own the intent detection and classification models for the bank's call-centre platform: labelling, training, and inference over live call transcripts from AWS Transcribe, across thousands of daily banking calls.
- Built an LLM ensemble labelling pipeline (GPT-4o mini, GPT-4.1 mini, GPT-5 mini) with consensus voting and escalation to larger thinking models for tie-breaks, to generate training labels for fine-tuning.
- Ran a knowledge distillation workflow using the ensemble as teacher to fine-tune Arctic Embed 2.0 Large into a domain student model for banking intent semantics.
- Built the continuous training pipeline on S3 and Apache Airflow: dataset ingestion, retraining, evaluation, and deployment to OpenShift.
- Benchmarked SVM (RBF), logistic regression, SetFit (head and full fine-tune), centroid, and fine-tuned Arctic Embed classifiers for production intent classification.
- Served Arctic Embed 2.0 Large with ONNX Runtime on CPU-only OpenShift, tuned with AVX-512 and related flags (no VNNI), inside a Spring Boot service on Java 25 that handles thousands of concurrent advisor and client conversations in real time.
- Trained a class-balanced logistic-regression head on the ensemble-labelled data to classify caller intent live as each conversation progresses, reaching up to 85% accuracy.

Tech stack: Java 25 · Spring Boot · Python · ONNX Runtime · Arctic Embed 2.0 Large · AVX-512 · scikit-learn · SetFit · SVM · GPT-4/4o/5 · Claude · S3 · Apache Airflow · OpenShift · AWS Transcribe/Bedrock

## Career History

- **Staff Software Engineer** · Vulnerability Management Group (VMG), Royal Bank of Canada (Jul 2024 to Sep 2025)
- **Lead Software Engineer** · SRE & Chaos Engineering, Royal Bank of Canada (Aug 2022 to Jul 2024)
- **Senior Software Developer** · Developer Experience & OSPO, Royal Bank of Canada (Nov 2020 to Aug 2022)
- **Software Developer** · Developer Experience, Royal Bank of Canada (Aug 2019 to Nov 2020)
- **Software Engineering Intern** · 3 terms (QE, Dev, SRE), Royal Bank of Canada (May 2017 to Aug 2019)
- **Full-Stack Developer** · Contract, University of Toronto (Sep 2017 to Feb 2018)

## Key Projects

- **Live Intent Classifier**: An LLM ensemble labelling pipeline (GPT-4o mini, GPT-4.1 mini, GPT-5 mini with consensus voting) labels call data, which trains a class-balanced logistic-regression head over Arctic Embed 2.0 Large embeddings. Served with ONNX Runtime on CPU-only OpenShift (AVX-512, no VNNI) in a Spring Boot service on Java 25, classifying caller intent in real time as the conversation progresses. | Metrics: Up to 85% accuracy, Thousands of live conversations, CPU-only, AVX-512 · Internal/proprietary
- **Vulnerability Triage Platform**: Centralized platform for vulnerability triage and exemption services across RBC subsidiaries in Canada, the US, and the UK. Ingests Snyk, NexusIQ, Kenna, Tenable, and Aqua into a Snowflake warehouse with triage dashboards. | Metrics: Canada · US · UK, Snowflake warehouse, Five scanner feeds · Internal/proprietary
- **Chaos Engineering Platform**: Automated large-scale chaos experiments across AKS, OpenShift, and VMware clusters using the Tanium and Gremlin APIs. Gives teams self-service chaos scheduling for VM and Kubernetes workloads. | Metrics: 90% faster setup, AKS · OpenShift · VMware, Self-service scheduling · Internal/proprietary
- **MirrML**: Flask app on the Clarifai image recognition API. Classifies clothing style (business, casual, evening) with a neural network trained on scraped image data, then matches users with friends who dress similarly. · [https://github.com/samhaque/MirrML](https://github.com/samhaque/MirrML)
- **LendR**: Android app that uses NFC for social micro-financing. A Karma system rewards on-time repayment; negative karma lowers borrowing limits. The backend tracks transactions and karma ratings. · [https://github.com/samhaque/FinTech_LendR](https://github.com/samhaque/FinTech_LendR)
- **HackTheValley API**: Production event management API for HackTheValley, a University of Toronto hackathon. Built with Go for high performance and deployed for the annual event. · [https://github.com/hackthevalley/htv-api](https://github.com/hackthevalley/htv-api)

## Skills

- **AI / ML Systems**: LLM Ensemble Labelling, Knowledge Distillation, Arctic Embed 2.0, Embedding Fine-tuning, ONNX Runtime, SetFit, SVM (RBF), Intent Classification, Semantic Clustering, RAG, Prompt Engineering, RASA NLU, Clarifai
- **Languages**: Python, Java, Go (Golang), JavaScript, TypeScript, SQL, Shell/Bash, C, C++, Ruby
- **MLOps & Data Pipelines**: Apache Airflow, S3 Data Pipelines, Continuous Training, Model Evaluation, Snowflake, Apache Kafka
- **Cloud & Containers**: OpenShift (OCP), Kubernetes, Docker, Azure AKS, AWS (Transcribe, Bedrock, S3), PCF
- **Frameworks**: Spring Boot, FastAPI, Flask, Django, React, Gin, ASP.NET Core, RabbitMQ
- **Databases**: PostgreSQL, SQL Server, Elasticsearch, MongoDB, Redis, MariaDB, MySQL, Oracle DB
- **DevOps & CI/CD**: GitHub Actions, Jenkins, Ansible, HashiCorp Vault, Artifactory, SonarQube, IBM UDeploy, Git
- **SRE & Observability**: Gremlin, Tanium, PagerDuty, ServiceNow, ELK Stack, Prometheus, Grafana, Moogsoft, Dynatrace
- **Security**: Snyk, NexusIQ, Kenna, Tenable, Aqua, Recorded Future, mTLS, PingFederate
- **AI Dev Tools**: Claude, ChatGPT / GPT-4o, GitHub Copilot, Cursor

## Education

- **Honours B.Sc., Statistics** · University of Toronto (2015 to 2021). Statistics major. Double minor in Geographic Information Systems (GIS) and Psychology.

## Volunteering

- **VP of Development** · AMACSS (Association of Mathematical & Computer Science Students) (Jun 2018 to Apr 2020): Led development projects and technical initiatives for the math and computer science student association at UTSC.
- **Developer / SRE** · Computer Science Enrichment Club (CSEC, University of Toronto) (Aug 2016 to Apr 2020): Ran club infrastructure and development projects, and mentored students in software engineering and SRE basics.
- **Web Developer** · Women in Computer Science, Statistics & Mathematics (WiCSM) (Aug 2020 to Present): Built and maintained the group's website. The group advocates for gender diversity in STEM at the University of Toronto.

## Full Version

For complete details: https://www.samiulhaque.com/llms-full.txt