Skip to content
karim.semaan(open to work)
WorkExperienceAboutSkillsContactResume ↓
Founder & CEO · AI / ML Engineer

I build machine learning models and the products that ship them.

Founder & CEO of SSS Tech · MS in AI at Northeastern. I take models from notebook to production.

View work →Get in touch
  • Measured calibration study · public PDF ↗︎
  • 7-model ensemble + 10k-run Monte Carlo
  • Automated payroll · FMCG/pharma
  • AWS ML – Specialty

Open to AI/ML roles + projects

Measured: 30/30 golden · 16/16 held-out→
MeasurementLLM CalibrationAgent-Tool Eval

Explore by area

Tap a lobe to explore

Pick an area to light up the brain and list its work.

MeasurementLLM CalibrationAgent-Tool Eval
01 / Selected Work

Things I've built.

Production client systems, applied-ML projects, and live in-browser demos. Public work runs in-page or links out; protected client work is previewable on synthetic data, and source and sensitive details stay private.

Start here

The two that best show how I work, end to end.

Measured:30/30 golden·16/16 held-out·Bastion 10/20 strict, 83.1% recall·reproduce it yourself ↗︎

  • KickCast preview
    01
    Graduate projectLive

    KickCast

    Turns a 21,371-match feature matrix into 10,000-run Monte-Carlo odds for every one of the 104 games of the 2026 World Cup: full probability distributions, not a single guess.

    An end-to-end ML system that predicts international football matches as a 3-class (home/draw/away) problem, then simulates the full 48-team 2026 World Cup from those probabilities. Seven classifiers plus a custom PyTorch net, SHAP explainability, a follow-up calibration study, and a live dashboard.

    • Python
    • scikit-learn
    • XGBoost
    • LightGBM
    • PyTorch
    • Optuna
    • 1.347 → 1.093 (calibrated)Log-loss (holdout)
    • 21,371 matches × 38 feat.Feature matrix
    • 7 classifiers + PyTorch netModels
    Open live↗︎View source↗︎
  • Bastion preview
    02
    Protected

    Bastion

    Compresses a 10–12-week security engagement into one workflow: a multi-stage Claude pipeline (an estimated ~80% token reduction) where every finding cites the evidence it came from and a human signs off.

    A production multi-tenant SaaS that runs the full cybersecurity-maturity assessment lifecycle (evidence vault, a multi-stage Claude gap-analysis pipeline against frameworks like NIST CSF 2.0, scoring, and client-ready reports) with a strict RLS-enforced consultant/client split. Live as an invite-only deployment, with a walkthrough available on request.

    • Next.js 15
    • React 19
    • TypeScript
    • Supabase
    • Claude (Sonnet + Haiku)
    • Pinecone (RAG)
    • 5Frameworks · NIST / CIS / ISO / SOC 2 / CMMC
    • cites evidence + human sign-offEvery finding
    • Supabase RLS (client/internal)Tenant isolation
    Private · request access

More flagship work

  • INTELIPS

    LLM-labeled email-priority NLP

    • Python
    • PyTorch
    • scikit-learn
    Completed

  • PayStream

    Enterprise payroll ETL

    • React
    • TypeScript
    • Vite
    Protected

  • Sol-Sniper

    AI Solana trading bot (MCP)

    • Python
    • LightGBM
    • scikit-learn
    Protected

  • Agrovio

    AgTech produce marketplace

    • Next.js
    • React
    • TypeScript
    Protected

More work

Showing 6 of 15 projects.

  • Kirby

    grounded portfolio assistant · guard-first · measured

    • LangGraph
    • Llama 3.1 8B
    • NVIDIA NIM
    Live

  • KickCast Calibration Study

    isotonic recalibration · 2022 WC holdout

    • Python
    • scikit-learn
    • XGBoost
    Study

  • Calibration Lab

    client-side probability-calibration explorer

    • TypeScript
    • Next.js
    • React
    Live

  • LLM Confidence Calibration

    are LLMs calibrated when they sound sure?

    • TypeScript
    • NVIDIA NIM
    • MMLU
    Study

  • Eval Gauntlet

    LLM regression testing · in-browser

    • TypeScript
    • React
    • Next.js
    Live

  • SEM BM

    Byzantine-chant learning games

    • Next.js 15
    • React 19
    • TypeScript
    Live
open source & beyond

Every repository.

The full inventory behind the curated work above: 48 repositories. 9 public repositories on GitHub; 8 shown here (two halves of the same course project are listed once). Forks and duplicate facets of the same project are not listed.

Showing 4 of 48 repositories.

  • kickcast-worldcup

    World Cup match prediction: leakage-safe 21,371-match pipeline, 7-model ensemble + PyTorch net, calibration-first evaluation, 10k-run Monte Carlo simulation

    Jun 2026Completed↗︎

  • PAEPS

    LLM-labeled email-priority NLP

    Jun 2026Completed↗︎

  • Achievements

    The source of this site: a Next.js + Supabase portfolio with typed project content, CI honesty gates, and the Kirby chatbot.

    Jun 2026PrivateWIP

  • Birthday_Buddy

    AI WhatsApp birthday concierge

    Jun 2026PrivateWIP
02 / Experience

Where I've worked.

A cloud / DevOps foundation (Harvard, Orna Therapeutics, Vocadian), then an AI Solutions Lead role shipping production AI for enterprise clients, now founding SSS Tech, an applied AI studio.

  1. Jun 2026 – Present

    SSS Tech

    Founder & CEO

    Boston, MA

    • Founded an applied AI studio that designs and ships production-grade AI and large language model systems, from a first notebook to a deployed product: data and architecture, multi-agent pipelines, evaluation and guardrails, cloud infrastructure, and the interface users touch.
    • Recent work includes Bastion, a multi-tenant platform that runs cybersecurity maturity assessments through a multi-stage AI pipeline scored against NIST CSF 2.0, CIS, ISO, SOC 2, and CMMC, and Birthday Buddy, an AI WhatsApp concierge that plans and runs personal celebrations end to end.

  2. Jun 2025 – Jun 2026

    (remote, part-time)

    BDO

    AI Solutions Lead

    Achrafieh, Lebanon

    • Built an automated payroll platform now used daily by Fortune 500 FMCG/pharma payroll teams for reporting (the PayStream work shown above).
    • Rolled out a real-time time-reporting & analytics system for 350+ employees, cutting reporting time ~80% (the TimeSheet platform).

  3. Sep 2024 – Jun 2025

    Vocadian

    DevOps Engineer

    Cambridge, MA

    • Migrated secrets to AWS Secrets Manager for SOC 2 compliance, closing off credential-leak risk.
    • Built CI/CD with GitHub Actions (~50% faster deploys) and migrated 10+ databases to RDS with zero data loss.

  4. Jan – Jul 2023

    Orna Therapeutics

    AWS Cloud / DevOps Engineer · Co-op

    Watertown, MA

    • Built AWS Lambda data-validation pipelines that cut data discrepancies ~15% for cleaner reporting.
    • Optimized GitLab CI/CD for Kubernetes and automated infrastructure with Terraform.

  5. Jan – Jul 2022

    Harvard University

    Cloud & IT Analyst · Co-op

    Cambridge, MA

    • Resolved 3,000+ ServiceNow tickets and 600+ Active Directory incidents.
    • Wrote PowerShell automation to deploy 1,000+ machines.
Live demos

Machine learning you can poke at.

Two interactive toys I wrote from scratch (no ML library), running in your browser. Train a small neural network and watch its decision boundary form, or race four optimizers down a loss surface. Real code, not screenshots.

Blue = class 0 · magenta = class 1 · the pale band is the decision boundary the network is still unsure about.

Epoch
0
Loss
–
Accuracy
0%
Dataset
Hidden units: 8 ×2 layers
Learning rate: 0.30
Activation
Calibration Lab →Score probabilistic predictions (log-loss, Brier, ECE), draw the reliability curve, and fit Platt or isotonic recalibration. Fully client-side, on synthetic illustrative data.Open the lab →
03 / About

Four years, one direction.

whoami

$ whoami

karim_semaan

$ cat role.txt

Founder & CEO · AI / ML Engineer

$ cat focus.txt

applied ML · LLM systems · ship to production

$ cat status.txt

open to ai/ml roles + projects

$ cat location.txt

Boston, MA · Beirut, LB

$ cat proof.txt

16/16 held-out · 30/30 golden · Bastion eval ↗︎

I'm Karim, an AI/ML engineer. I started with a Flask app in 2022 and a stubborn curiosity about how things actually work. That turned into a CS foundation (algorithms, OOD, the Berkeley Pac-Man AI projects), then AWS cloud certifications, then graduate-level Deep Learning, NLP, and Reinforcement Learning.

Today I do the part I love most: taking models out of the notebook. Whether it's an ensemble predicting a World Cup or an LLM reasoning over security controls, I care about the whole path: the math, the data, and the product people end up touching. I've also shipped production systems for clients (security, payroll, and a produce marketplace), which appear here as anonymized previews on synthetic data.

Now I run that work as SSS Tech, an applied AI studio I founded to design and ship production-grade AI and LLM systems from a first notebook to a deployed product. I'm building on the AWS Machine Learning – Specialty certification and an M.S. in Artificial Intelligence (ML concentration) at Northeastern University, while shipping more of this work as runnable, public demos.

“A model is only as good as the product that delivers it.”

Beyond the desk: chess, tennis, basketball, ping-pong, badminton, skiing. I track my lifts with apps I built, study my League of Legends matches with my own tracker, and practice Byzantine chant notation with a web app I wrote for it.

04 / Timeline

How I got here.

  1. 2026Founded SSS Tech, an applied AI studio shipping production AI & LLM systems. Shipped flagship products: KickCast (ML prediction) and Bastion (LLM-powered security).
  2. 2025AWS Machine Learning – Specialty certified. Graduate AI/ML: Deep Learning, NLP, ML Foundations, Algorithms; first shipped automation tools.
  3. 2024AWS Solutions Architect – Associate certified.
  4. 2023CS fundamentals: Java OOD, web (React/Node), Berkeley Pac-Man AI (search, multiagents, RL). First GPT experiments.
  5. 2022First builds: Flask + databases.
05 / Capabilities

What I work with.

ML / Deep learning
PyTorch, scikit-learn, XGBoost / LightGBM, Ensemble methods, CNNs / ResNet, Transformers / BERT, Reinforcement learning, Monte Carlo simulation, Optuna, SHAP
LLM / GenAI
Claude API, OpenAI GPT-4o, NVIDIA NIM, Groq, LangGraph, RAG (Pinecone), ReAct agents / tool-use, Agentic / multi-agent (MCP), Prompt engineering, Injection defense & guardrails
Calibration & evals
Model calibration (isotonic / Platt), ECE / Brier / log-loss, Reliability diagrams, LLM evals (golden + held-out), Agent evaluation, Precision / Recall / F1, Failure-mode taxonomies, Reproducible studies
Data & analysis
Pandas, NumPy, SciPy, OpenCV / Pillow, Matplotlib / Plotly, Jupyter, Feature engineering, ETL / data pipelines, openpyxl
Ship it
Next.js, React / React Native, TypeScript, Tailwind / shadcn, Supabase / PostgreSQL, FastAPI / Flask, Docker, Vercel, npm / OSS
Languages
Python, TypeScript, JavaScript, Java, SQL, Racket, Common Lisp
Credentials

Certified by AWS.

  • AWS Certified Machine Learning – Specialty

    Amazon Web Services Training and Certification

    Issued
    January 17, 2025
    Expires
    January 17, 2028
    AdvancedVerify ↗︎

  • AWS Certified Solutions Architect – Associate

    Amazon Web Services Training and Certification

    Issued
    July 30, 2024
    Expires
    July 30, 2027
    IntermediateVerify ↗︎
Education

Academic foundation.

  • Northeastern University

    M.S. Artificial Intelligence, Machine Learning concentration

    Focus · Deep Learning · Natural Language Processing · Reinforcement Learning

    Prior · B.S. Computer Science & Mathematics · B.S. GPA 3.70 · Dean's List · NU Partners Scholarship

    Boston, MA · Sep 2025 – May 2027 (expected)

    Graduate
  • Graduate & upper-division coursework

    Deep Learning, Natural Language Processing, Machine Learning, Artificial Intelligence, Reinforcement Learning, Algorithms, Theory of Computation, Object-Oriented Design

06 / Contact

Let's put a model in production.

Open to full-time AI/ML roles and freelance projects. The fastest way to reach me is email, or just send a note below.

semaankarim02@gmail.com
LinkedIn ↗︎GitHub ↗︎Resume ↓

Your note goes straight to my inbox. No newsletter, no spam.

Measured:30/30 chatbot evals·20-case Bastion eval·Serving·Calibration Lab

© 2026 Karim SemaanBuilt with Next.js, Tailwind & Supabase.LinkedIn ↗︎GitHub ↗︎