Naim

Lead AI & Computer Vision Engineer

Sep 2025 - Dec 2025 • 3 mos

Architected "Advanced Security & Attendance Ecosystem": A Real-Time Computer Vision Suite I designed a production-grade surveillance and management system aimed at automating physical security, identity management, and HR operations. This full-stack application integrates state-of-the-art computer vision models with complex backend logic. What I Built: Dynamic Real-Time Recognition: Developed 'Verify Mode' (Face/QR/RFID) and 'Watch Mode' (YOLO+DeepSORT). Engineered a dynamic update engine where transferring IDs between tabs (e.g., Unknown → Personnel) results in instant recognition updates across live cameras without system downtime. Smart Vault & History Recovery: Built a robust Identity Vault system capable of restoring deleted users. The system automatically retrieves and re-links complete historical data (attendance logs, previous metadata) from the vault upon recovery, ensuring zero data loss. Automated Operations Engine: Engineered complex backend logic for Smart Payroll (calculating Net Pay/OT/Fines) and included 'Manual Override' tools for admins to correct incomplete attendance logs with audit trails. Proactive Threat Intelligence: Integrated an SMTP alert system that triggers instant emails with high-res snapshot evidence when "Suspicious" blacklist targets or "Unknown" faces are detected in restricted zones. System Health & Analytics: Designed a live dashboard monitoring Hardware stats (CPU/FPS) and implemented an Auto Daily Digest that compiles and emails a comprehensive summary of daily logs at a scheduled time. Project Verdict: This ecosystem bridges the gap between raw AI models and actionable business logic. It demonstrates my ability to build end-to-end secure systems where real-time biometric tracking seamlessly integrates with GDPR-compliant data management and automated HR workflows.

AI Engineer (RAG & Local LLMs)

Jul 2025 - Sep 2025 • 2 mos

Engineered "Smart Document Analyst": An Offline AI for Secure Document Intelligence I architected a production-ready, privacy-first RAG application designed for high-stakes analysis without cloud dependencies. Unlike standard wrappers, this system runs 100% offline on consumer hardware (Intel i5/12GB RAM), ensuring zero data leakage for sensitive legal or research documents. What I Built & Optimized: Resource-Efficient Architecture: Successfully deployed Mistral-7B-Instruct-v0.1 (Q5_K_M) for reasoning and all-MiniLM-L6-v2 for semantic retrieval. Optimized the entire pipeline to function smoothly on limited hardware without requiring high-end GPUs. Hallucination-Free Q&A: Implemented a strict "retrieval-only" mechanism where the AI answers solely based on provided documents, supporting every claim with verifiable inline page citations. Smart Analysis & Translation: Deep Reasoning Engine: Engineered a multi-step "Chain-of-Thought" logic for Novelty Detection and Critical Reviews. This allows the AI to "think" in stages before generating an output, ensuring human-level analytical depth. Offline Translation: Integrated M2M100-418M to provide secure translations across 10+ languages directly within the analysis pipeline. Rigorous "LLM-as-a-Judge" Audit: I subjected the system to a transparent audit by ChatGPT-5 and Gemini 2.5 Pro. The verdict? It achieved 9.0/10 in Q&A Accuracy and 9.2/10 in Critical Review, validating that optimized local models can rival cloud APIs. Performance Note: Please note that inference speed is limited as the system runs entirely on a local CPU (Intel i5) to guarantee 100% air-gapped privacy. While slower than GPU-based cloud APIs, this architectural trade-off ensures zero data leakage while maintaining deep, multi-step analytical precision.