Applied AI Researcher · AI Engineer · Software Engineer
I build practical AI systems, research-driven software, and automation tools that connect machine learning, trustworthy evaluation, and real-world business impact.
Explore the code, inspect the recorded results, or help improve a project.
| Project | What you can explore | Current boundary |
|---|---|---|
| Biomedical Hybrid IR | Compare lexical, dense, hybrid, and reranked biomedical search on NFCorpus. Results · Reproduce | Recorded six-method study; portal displays exported results rather than live model inference. |
| DriftCVE-NLP | Investigate vulnerability classification and retrieval as language changes over time. Dataset · Reproduce | Recorded chronological experiments; live inference is pending. |
| Drift-Robust TinyML | Study electronic-nose sensor drift, compact models, and resource-aware explanations. Protocol · Reproduce | Host-side experiments and numerical equivalence are documented; physical MCU validation remains blocked. |
Research portals: Biomedical IR · DriftCVE-NLP · TinyML
- Lead.AI — lead capture and business automation across web and mobile; an evolving MVP.
- JavaScript 0 → Hero — interactive lessons, coding practice, and device-local learning progress.
- SitaRam — multilingual study workflows with source provenance and corpus approval gates.
Explore the full portfolio, materials, and project roadmaps →
I welcome developers, researchers, and people who can help test practical workflows.
| Interested in… | A useful way to participate | Where to begin |
|---|---|---|
| Biomedical search | Reproduce a baseline, investigate a failed query, or improve setup instructions. | IR issues |
| Cybersecurity NLP | Review preprocessing, reproduce temporal evaluation, or report dataset problems. | DriftCVE issues |
| Embedded ML | Review export workflows or discuss real-board validation and instrumentation. | TinyML issues |
| Interactive learning | Try a lesson or challenge and report its steps, expected behavior, and actual outcome. | JavaScript learning issues |
Read the project's setup and contribution guidance, check existing issues, and propose a focused change before beginning a large implementation. Reproduction reports are welcome even when your results differ.
For research collaboration or engineering opportunities, email me.
- Reproducible experiments: explicit data provenance, versioned configurations, and saved artifacts.
- Careful evaluation: report measured findings, limitations, and pending work with their supporting evidence.
- Practical delivery: documented setup, automated checks, and complete user workflows.
Core tools: Python · scikit-learn · Hugging Face · TypeScript · React / Next.js · FastAPI · Flutter · GitHub Actions
Research background, full project evidence, and technical toolkit
I am pursuing a PhD in Computer Engineering at Florida Atlantic University, in the Department of Electrical Engineering and Computer Science. I hold an MS in Computer Science with an AI/ML focus from Westcliff University.
My research connects distribution shift, explainability, and resource-aware machine learning. I build reproducible experiment pipelines and interfaces that make their evidence accessible. My applied engineering work spans AI business automation, web and mobile applications, and software testing.
| Project | Research question and implementation | Current evidence |
|---|---|---|
| Drift-Robust TinyML | How do compact electronic-nose models and explanations behave under chronological sensor drift? UCI data, frozen evaluation protocols, XAI, and embedded equivalence checks. | Recorded baseline, drift, and host-side XAI experiments; fused FP32 inference and local-XAI equivalence. Physical MCU latency, memory, and energy validation remain blocked. |
| DriftCVE-NLP | How does vulnerability language change over time, and what happens to CVE classification and retrieval? NVD ingestion, chronological splits, lexical and embedding models, provenance, and result export. | Versioned dataset and recorded experiments with confidence intervals and per-example predictions. Live inference remains pending. |
| Biomedical Hybrid IR | When do lexical search, dense retrieval, hybrid fusion, and reranking help biomedical search? TF-IDF, BM25, BGE, MedCPT, reciprocal rank fusion, and cross-encoder evaluation on NFCorpus. | Six evaluated methods, result tables, statistical comparisons, query-level error analysis, and reproducibility documentation. A completed course study with a working manuscript. |
| DriftGuard-IoT | Can leakage-safe intrusion detection remain useful across IoT/IIoT datasets and changing traffic? Baseline reproduction, drift monitoring, adaptation, and resource evaluation. | Development runs and tested evaluation utilities; real-data research campaign remains blocked. Synthetic runs are non-reportable. |
| Customer Behavior Prediction — Part 2 | How should financial prediction be evaluated beyond static accuracy? Temporal stability, calibration, explanation stability, and cost-sensitive decisions. | Research protocol, preflight/provenance runner, and evaluation utilities. Empirical manuscript results await benchmark acquisition and execution. |
Explore the research portals: TinyML · DriftCVE-NLP · Biomedical IR
Portals communicate experiment artifacts and project status; deployment alone does not establish model validity or hardware performance.
| Project | Engineering focus | Stage |
|---|---|---|
| Lead.AI Product Platform | Lead capture and business automation across FastAPI, React, Flutter, and Firebase. | MVP; production-readiness work continues. |
| Lead.AI Fraud Benchmark | Reproducible tabular training, held-out evaluation, deterministic reasoning tasks, artifact checksums, and controlled publishing workflows. | Controlled synthetic benchmark; results do not establish real-world fraud performance. |
| SitaRam | Multilingual Flutter application with corpus provenance, content validation, and evidence-grounded assistance. | Application and release workflows; source coverage remains incomplete. |
| Working Woman Report | Next.js editorial platform connecting a canonical weekly story with publishing, archive, and distribution workflows. | Rebuild with content schemas and provider adapter foundations. |
Additional operational prototypes: FleetDesk and DailyOps are documented in the portfolio roadmaps. Their source repositories currently require collaborator access.
Additional work: payment management, AI companion interfaces, construction lead management, and JavaScript learning tools.
- Reproducibility: versioned configurations, explicit data provenance, chronological splits, and saved result artifacts.
- Honest evaluation: distinguish measured findings, synthetic demonstrations, pending experiments, and blocked hardware work.
- Useful explanations: assess explanation fidelity, stability, and computational cost alongside predictive performance.
- Reliable delivery: automated tests, validation gates, documented setup, and clear application boundaries.
| Area | Tools and methods |
|---|---|
| Research and ML | Python, scikit-learn, Hugging Face, SHAP, Jupyter/Colab, statistical evaluation |
| NLP and retrieval | TF-IDF, BM25, embeddings, transformer models, rank fusion, cross-encoder reranking |
| Application engineering | TypeScript, React, Next.js, FastAPI, Flutter, Firebase, SQLite |
| Testing and delivery | pytest, Selenium, Playwright, GitHub Actions, Git, Vercel |
| Embedded research | C/FP32 export, numerical equivalence, resource-aware XAI; nRF52840 hardware validation planned |
Models and datasets · Lead.AI Labs · Portfolio
Profile presentation updated October 6, 2026. Project status summarizes repository documentation; consult the source artifacts for current evidence.