hi there, I'm
18:18:24
Mohammad Anzar Akhtar
Most teams are not short on ideas. They are short on engineers who bridge research and production. I ship projects end-to-end, from PyTorch deep learning pipelines to security-hardened, responsive Next.js web applications.
Work Experience.
Infosys
Full Stack Development Intern (Virtual)
Shipped EcoPackAI, an AI packaging recommendation engine scoring materials on cost, carbon emissions, strength, and recyclability. Trained Random Forest and XGBoost predictive models on engineered data, and built Flask REST APIs backed by MySQL and a responsive interface.
Projects.
CampusConnect — Full-Stack Platform
- Architected and shipped a 29-route collaboration platform for college students with real-time feeds, DMs, marketplace, AI assistant, and coding arena on Next.js 15 & Supabase.
- Led security-hardening pass: college-scoped Row-Level Security policies, server-side OTP/email validation via Edge Functions, and signed URLs for private storage.
- Eliminated duplicate-vote race conditions via atomic RPCs and built motion-driven UI interactions using React Three Fiber and GSAP.
CI-MRGS: Multi-Region Glioma Segmentation
- Built a deep learning framework for multi-class brain tumor segmentation on BraTS 2024 using multi-modal MRI input (T1, T1c, T2, FLAIR) with Attention U-Net.
- Engineered a production-style PyTorch training pipeline with automatic checkpointing, mixed-precision (AMP), EMA, TensorBoard logging, and Kaggle GPU resumption.
- Designed a hybrid Focal Tversky + Boundary Loss and augmentation strategy to counter severe tumor sub-region class imbalance; prepared manuscript in IEEE format.
AIOS — Personal AI Operating System & Agent Runtime
- Architected a 4-layer unidirectional OS (Core → Integration → Skill → Interface) with a kernel state machine and heuristic intent router executing system queries with 0 LLM hops (7.4 µs dispatch latency).
- Engineered an episodic agent memory system pairing Qdrant vector search with automatic SQLite cosine-similarity fallback during network partitions and MinIO (S3) for durable checkpoint recovery.
- Implemented a two-phase action approval protocol (/v1/actions/propose → confirm) with local Ollama inference budgeting (Qwen 9B, DeepSeek R1) and SSE real-time telemetry (/v1/events).
- Maintained 100% test pass rate across 1,200 unit and integration tests (pytest in 3.02s) with automated Bandit security verification.
Skills
Education.
IILM University, Greater Noida
2024 - 2028Bachelor of Technology - B.Tech, Computer Science & Engineering
Certifications & Specializations
2025Oracle DBMS Certified (SQL & PL/SQL) · Infosys Springboard Data Science
Let's work together.
I'm always interested in new opportunities and exciting engineering challenges. Whether you have a project in mind or want to discuss AI systems, I'd love to connect.
Currently available for AI/ML engineering, full-stack internships, and new-grad roles
Response time: Usually within 24 hours