Mohamed Martini
NH · 617-955-4276 · leverai@leverai.tech · linkedin.com/in/mohamed-martini · https://leverai.tech
Skills: Python, AI/ML (LangChain, LangGraph, TensorFlow, PyTorch), Cloud (GCP - Cloud Run, CloudSQL, Cloud Trace), Data (PostgreSQL, BigQuery, Apache Beam), DevOps (Docker, CI/CD), Web (FastAPI), Flutter.
PROFESSIONAL EXPERIENCE
Founder & Principal Engineer | Lever AI Full-time · July 2025 - Present
- Founded a consulting and SaaS development LLC specialized in AI-powered, data-driven software solutions.
- Showcase Project: Guesstimate: Not Trivia! (Google Play, App Store)
- Designed and shipped a full-stack social estimation game to Android and iOS, featuring real-time multiplayer via Firestore, global leaderboards, and freemium monetization (RevenueCat, Unity Ads).
- Built an LLM-powered ETL pipeline using smart seed selection and pgvector embeddings for semantic deduplication, generating 700+ high-quality, diverse, and fact-checked questions at <1 cent per question.
- Architected real-time, fair question assignment for multi-player rounds using optimized SQL.
- Implemented end-to-end distributed tracing with OpenTelemetry, correlating Flutter frontend events to FastAPI backend spans via Cloud Trace and Cloud Logging.
- Established GitFlow CI/CD pipelines with automated testing, and dev/prod building and deployment.
- Leveraged AI coding agents to accelerate cross-platform Flutter development while maintaining full architectural oversight of the Python backend.
Machine Learning Engineer III | Pison Technology Full-time · Oct 2022 - July 2025 · 2 yrs 10 mos
- Promoted from MLE to MLE III in April 2025.
- Ran 1,200+ TensorFlow experiments for sleep stage classification, improving the baseline F1-score by 15%. Developed two custom loss functions to address class imbalance. Implemented a custom attention layer for feature mining, improving performance while reducing the size of the feature set by an order of magnitude.
- Developed real-time biosignal gesture detection models. Wrote Cythonized extension modules, and vectorized bottleneck functions, significantly improving the speed of online prediction and offline modeling.
- Led a "Hackaweek" project with two colleagues to implement a state-of-the-art domain-invariant classification network with adversarial training in PyTorch, following the approach outlined in its original manuscript.
- Led the adoption of BigQuery as a data warehouse solution to streamline the handling of analytics data.
- Deployed and monitored streaming and batch ML models using Google Model Registry and Cloud Functions.
- Wrote a scalable ETL pipeline using Apache Beam which brought down the compute time of 56 GB of data from 2 days to 15 minutes at a cost of 25 cents.
- Standardized the Python codebase, CI/CD workflows, and dependency management, enabling seamless collaboration between colleagues across operating systems.
- Maintained cross-team communication to ensure consistency and optimal integration of ML solutions.
Research Assistant | UNIVERSITY OF MASSACHUSETTS LOWELL Jan 2021 – June 2022
- Implemented Reinforcement Learning agents for cooperative multi-agent Search and Rescue simulations.
- Implemented object detection training and inference pipelines. Used Generative AI for data augmentation.
EDUCATION
- M.S. Computer Engineering · GPA 4.0 · UMass Lowell, 2022
- B.S. Electrical Engineering · GPA 3.8 · UMass Lowell, 2021
REFERENCES
References available upon request.