PhD Student | Senior Backend Engineer | AI Research Enthusiast | Software Architecture Designer
I'm a first-year PhD student at the intersection of agentic AI systems, software architecture, and energy-efficient AI. My research focuses on designing green, trustworthy, human-centered agentic AI systems that can reliably design software architectures.
My PhD work addresses critical gaps in how agentic AI systems are designed and evaluated:
- Architectural Topologies & Task Mapping – How do agentic AI systems decompose software architecture design tasks, and what orchestration patterns are most effective?
- Engineer Role Redefinition & HITL Patterns – How should human-in-the-loop integration reshape the role of software architects when working with agentic systems?
- Vulnerabilities & Defense Mechanisms – What are the failure modes of agentic orchestrators (especially in smaller language models), and how can we build resilient systems?
- Green Architecture Tactics & Sustainability Metrics – How can we measure and optimize energy consumption across agentic AI workflows?
- Plan Mode & Architecture Detection – How do agentic coders understand domain structure?
- HLD/LLD Generation – Bridging high-level design intent to low-level code
- Multi-Agent Frameworks:
- Tree-sitter-based symbol graph extraction
- PageRank over code property graphs
- Agent orchestration patterns (ReAct, Tree-of-Thought variants)
- Budget-forcing and constrained CoT execution
- Zero-Think, LessThink, Chain-of-Draft approaches
- Lost-in-the-Middle effect mitigation (Liu et al., TACL 2024)
- Context fill ratio optimization and energy-per-token metrics
- Phi-3 Mini, Gemma-2 2B, Qwen-1.5 evaluation
- Trade-offs between model size, latency, accuracy, and energy
- Benchmarking agentic system reliability at scale constraints
Languages & Frameworks:
- Python (NumPy, Pandas, scikit-learn, Langchain, LiteLLM)
- Java EE, Spring Framework
- Django, FastAPI
- C#, C++, Roff
Tools & Platforms:
- Git / GitHub
- Zotero (academic management)
- LaTeX & Markdown (research documentation)
- n8n (workflow automation)
- Docker & containerization
- CI/CD pipelines
AI/ML Specific:
- LLM APIs: Anthropic Claude, OpenAI, DeepSeek
- CodeCarbon, NVML (energy measurement)
- BM25, dense embeddings (retrieval)
- Structured output parsing (JSON, YAML)
- MCP (Model Context Protocol) development
- Education: PhD (Ongoing), Software Architecture & Agentic AI
- Professional Experience: Senior Backend Engineer with full-stack capabilities
- Instructional Roles: Teaching and mentoring in software engineering
- Open Source: Active contributor and maintainer of research tools
- Languages: Arabic (native), English (research & technical)
- LinkedIn: in/dr-mustafa-assaf
- Twitter/X: @MKASSAF
- GitHub: mkassaf
- Email: Research inquiries welcome on LinkedIn
"The future of agentic AI in software engineering isn't about raw reasoning power—it's about honest constraints, meaningful human integration, and measurable environmental impact."
- Energy as First-Class Constraint – Not an afterthought, but a design dimension
- SLM-First Thinking – Optimize for models that most engineers can actually deploy
- Human-Centered Orchestration – HITL isn't a fallback; it's core architecture
- Vulnerability-Driven Design – Build resilience by understanding failure modes
- Reproducible Benchmarking – Architecture-specific evaluation beats one-size-fits-all metrics
I'm always interested in:
- 🤝 Joint research on agentic AI and software architecture
- 🔍 Discussions on energy-efficient AI systems
- 📊 Benchmarking agentic AI frameworks
- 📚 Building curated resources for the community
- 🎓 Mentoring and knowledge transfer
Feel free to open issues, submit PRs, or reach out on LinkedIn for research collaborations.
Last Updated: June 2026 | PhD Year 1 | Research in Progress
"In every agentic system design decision, ask: How much energy? How much human judgment? How much risk?"