I’m Soroosh Esmaeilian, a Machine Learning Engineer with an MSc from the University of Calgary. Currently exploring LLMs, RAG systems, and tool-use agents while drawing on industrial experience shipping multi-task learning anomaly detection on enterprise IoT gateways.
From research to production. I’ve trained models that ran on enterprise edge devices and now build software that puts language models to work.
Machine Learning Engineer with an MSc in Computer Science from the University of Calgary. As part of a three-person team, I designed and deployed an multi-task learning anomaly detection model running on IoT gateway devices for enterprise clients.
These days I’m investigating LLMs and RAG systems: Claude and OpenAI APIs, pgvector embeddings, prompt engineering, and tool-use agents, while keeping current with the fast-moving AI and software engineering stack.
Comfortable across the ML toolkit (Keras, TensorFlow) and full-stack development (Next.js, FastAPI, PostgreSQL).
LLM-powered products I’ve designed, built, and shipped.
A cloud-native rebuild of Prima on Google Cloud. A five-node LangGraph fleet, reasoning with Gemini on Vertex AI, turns plain-English questions into read-only BigQuery queries over Alibaba’s 247M-sample cluster trace, scores anomalies (MAD/EVT), ranks the metrics behind them, and writes the briefing.
Visit site
An agentic server-health platform built on LangGraph. A fleet of Claude-driven agents writes and runs its own SQL over real server telemetry, detects anomalies, forecasts where a machine is heading, and grades its root-cause attribution against ground truth — then writes a plain-English reliability brief.
Visit site
Find organizations actually doing the R&D you care about across Canada, the US, UK, and Australia. It indexes their public research-funding history and matches it to your query, resume, or paper using hybrid vector + full-text search with an LLM reranker.
Visit site
Canada-wide rental search. You describe what you’re looking for in plain English; the system returns real rental listings on a map, with reasoning that cites each pick by id.
Visit site
A Next.js app that lets anyone discover connections between notable individuals by finding instances where they appear together in a photo, either directly or through a short chain of mutual photos. Inspired by six degrees of separation.
Visit siteFrom training neural nets to wiring up agents and APIs in production.
Research and applied ML, from packet-level simulators to enterprise IoT.
Collaboratively designed and developed EdgeInsight with a team of three, a multi-task learning anomaly detection model built in PyTorch for deployment on Litmus Edge, an IoT gateway trusted by enterprise companies.
Designed a network-wide sampling system for SDNs that adapts to dynamic flow rates. Packet-level simulations in ns-3.
Applied ML and signal processing to neuroimaging data (FNIRS, EEG) for cognitive workload and BCI classification.
Research focus: adaptive packet sampling in SDNs using ns-3 simulations.
Peer-reviewed work from my MSc research at the University of Calgary.
Open to ML engineering, LLM/RAG, and full-stack product roles.
I’m happy to chat about LLM architecture, RAG systems, anomaly detection, or full-stack engineering. Reach out and let’s talk.
Email me