@laminesn · AI Engineer · ML Systems · MLOps, Agents & LLMs
I build AI systems that are theoretically grounded and production-ready — from agentic LLM workflows to deep representation learning.
AI Engineer at BWXT and Graduate Researcher at the Florida Tech NETS Lab, working across LLMs, agentic systems, computer vision, and the cloud infrastructure that ships them.
I'm a software engineer and machine learning engineer focused on building intelligent, scalable, and practical AI-powered systems. My work spans applied research, model development, and the cloud infrastructure that ships them — from computer vision and NLP to agentic systems, MLOps, geospatial analytics, and AI workflow automation.
I care about systems that don't just demo well, but hold up in production: observable, reproducible, and honest about their limits.
A disciplined toolkit prioritizing correctness, observability, and shipping speed.
ACTIVEMulti-agent LLM assistant built on Google ADK + LangChain + FAISS, with a dual-LLM security pattern keeping RAG and user data fully local.
ImpactMLOps stack (MLflow + Ray Serve + Dagster) with RL-based personalization cut deploy time by 20%.
SHIPPEDGeospatial pipeline that identifies schools, hospitals, and fire stations exposed to FEMA-designated flood zones. Combines live FEMA NFHL and OpenStreetMap data with spatial operations, risk scoring, and an interactive Streamlit dashboard.
ImpactRanks critical infrastructure by flood proximity across Houston, TX; enables data-driven emergency planning and resource prioritization.
RESEARCHEnd-to-end pipeline aligning Gazepoint eye-tracking (~60 Hz) with video frames to train a ResNet-18 encoder–decoder predicting 853×480 attention heatmaps from gaze.
ImpactImproved localization error 27.2% over baseline (340.7px → 248.0px); 45.4% of predictions within 100px on held-out sequences.
SHIPPEDRepeatable, AI-trainable pipeline that turns raw 2–5 minute talking-head clips into broadcast-ready 45–60 second social media shorts. Trims dead air, normalizes audio to EBU R128, and burns in Whisper-generated captions.
ImpactPackaged as a verifiable tool-use task with pass/fail evaluation criteria, suitable for training AI agents inside real creative software.
RESEARCHReusable PyTorch module with a custom entropy loss and HSIC-based regularizer for analyzing and improving representation learning in convolutional networks.
ImpactPublished in Entropy (MDPI, 2026) — improves convergence and accuracy across CNN architectures.
SHIPPEDFull-stack web app for early COVID-19 screening via AI-powered cough analysis, with a dual-attention CNN (channel + spatial) over audio spectrograms.
Impact70% inference accuracy with real-time audio preprocessing; deployed on Google Cloud Run with JWT-secured Next.js dashboard.
Roles building ML infrastructure, research prototypes, and the platforms beneath them.
How deep networks organize information across layers, channels, and training time.
Latent mutual information and HSIC-based regularization for CNN training dynamics.
Multi-agent LLM orchestration, tool use, and autonomous workflows with ADK + LangChain.
Reproducible training, evaluation, and deployment of LLM-backed applications.
Attention modeling, gaze supervision, and CNN interpretability on real-world video.
Ray Serve, Dagster, MLflow, and CI/CD for honest, observable model delivery.
Local-first inference and dual-LLM security patterns for sensitive RAG workflows.
RL-based personalization and policy learning inside production AI systems.
Work at the intersection of information theory and deep representation learning.
Open to AI/ML Engineering, Software Engineering, and Research Engineering opportunities — full-time, contract, or research collaborations.
Recruiters, founders, engineers, researchers, collaborators — if you're working on something interesting at the intersection of software and AI, I'd love to hear from you.
lamineszn@gmail.com