Mohamed Amine Ghorbali

Designed and developed three production-oriented AI microservices for Hirevyn, an AI-powered HR/recruitment platform, using locally-hosted LLMs to meet strict data-privacy requirements.

• Built a hybrid RAG-based HR policy assistant combining BM25 sparse retrieval, dense embeddings, RRF fusion, and cross-encoder re-ranking, achieving 86.7% Recall@5 with citation-backed source passages.
• Designed an asynchronous resume analysis engine (FastAPI + Celery) that parses PDF/DOCX resumes into structured Pydantic schemas and generates candidate scoring, interview prep questions, and study flashcards via local LLM inference — reaching 87.4% field-level parsing accuracy and 95.8% JSON schema adherence.
• Developed a real-time voice interview agent over WebSocket/WebRTC, integrating faster-whisper ASR, Silero VAD, and Piper TTS with an LLM-driven question-generation state machine.

Working on a teleconsultation and patient follow-up platform, focused on full-stack serverless architecture and scalable cloud deployment.

• Implementing scalable backend services using AWS Amplify, Lambdas, and RDS with Django REST APIs.
• Building responsive and cross-platform UIs with React and Ionic for seamless patient and doctor interaction.
• Architected CI/CD pipelines and automated testing workflows to streamline deployment across staging and production environments.
• Refactoring legacy code to improve maintainability and performance.

Built a real-time sign language to text translation system using deep learning. Designed a two-stage pipeline combining visual feature extraction and sequence modeling for end-to-end gesture recognition.

• Engineered CNN architectures for precise spatial encoding of sign language video frames, achieving over 90% accuracy.
• Developed and optimized LSTM models for temporal sequence interpretation of gestures.
• Trained a gloss-to-text decoder using Transformer-based seq2seq models for fluent natural language generation.

Engineered the core backend infrastructure for an e-commerce platform tailored to game developers and digital artists, focusing on secure transactions, data integrity, and scalable API design.

• Architected a robust Laravel backend with RESTful APIs to support high-throughput operations for digital asset distribution and user management.
• Designed an optimized relational MySQL database schema for efficient querying of complex product catalogs, user profiles, and transactional records.
• Integrated secure payment processing gateways with robust webhook handling for real-time transaction verification and automated order fulfillment.

Designed and deployed an AI-powered 3D image analysis platform integrating a Laravel UI and FastAPI-based inference engine. Built a full pipeline for segmentation, classification, and cancer risk prediction.

• Used 3D U-Net with CBAM attention to segment pulmonary nodules in CT scans.
• Classified nodules with 3D CNNs into benign/malignant categories and subtypes.
• Predicted cancer stage, mortality risk, and lung function decline.

Fine-tuned a domain-specific LLM (LLaMA 2) to analyze SEC financial filings (10-K, 10-Q), extract structured financial metrics, identify trends, and produce executive summaries.

• Curated a dataset of annotated financial filings using the SEC EDGAR API and regex-based entity extraction.
• Achieved 91% extraction accuracy for revenue, net income, EPS, and related financial metrics.
• Generated MD&A section summaries rated 85%+ in clarity and factual correctness by domain experts.
• Enabled automatic trend detection and risk flagging through in-context financial language modeling.