name: Onirudda Islam
role: Co-Founder & CTO
companies:
- Silicornya Ltd (silicornya.com)
- MemeMates (mememates-website.web.app)
education:
- degree: BSc (Double Major)
school: Massachusetts Institute of Technology (MIT)
majors:
- Computer Science (Course 6-3)
- Physics (Course 8)
- degree: MSc
school: Massachusetts Institute of Technology (MIT)
field: Theoretical Particle & Nuclear Physics
experience:
- company: Google
title: Software Engineer L5
- company: Oracle
title: Cloud Engineer IC-3
current_research:
- Persistent memory architectures for AI agents
- Solving the context window limitation problem
- Long-horizon memory retrieval in LLM-based systems
all_repos: 🔒 Private (stealth mode)┌─────────────────────────────────────────────────────────────┐
│ 🧠 AI PERSISTENT MEMORY SYSTEMS │
│ │
│ Context Window ──► Memory Compression ──► Retrieval │
│ ↓ ↓ ↓ │
│ Token Limits Semantic Index Agent Recall │
│ │
│ Goal: Enable agents to maintain unbounded context │
│ across sessions through intelligent memory mgmt │
└─────────────────────────────────────────────────────────────┘
Active Problem: LLMs are constrained by finite context windows, making long-horizon reasoning and persistent agent state a fundamental bottleneck. I'm exploring hybrid approaches combining vector memory stores, hierarchical summarization, and episodic recall to give agents true long-term memory.
🔒 Note: All project repositories are private. Enable "Include private contributions" in your GitHub settings to reflect full activity.
| 🎓 Education | 💼 Experience | 🚀 Ventures |
|---|---|---|
| MIT BSc — CS (6-3) + Physics (8) | Google — Software Engineer L5 | 🏢 Silicornya Ltd — CTO & Co-Founder |
| MIT MSc — Theoretical Particle & Nuclear Physics | Oracle — Cloud Engineer IC-3 | 😂 MemeMates — Co-Founder |
Technology company pushing boundaries in AI infrastructure and software engineering. |
Social platform connecting people through humor and meme culture. |
📌 Persistent Memory in AI Agents
└─ How do we give agents genuine long-term memory beyond a context window?
└─ Hierarchical episodic memory + semantic compression
└─ Retrieval-Augmented Memory (RAM) as a first-class primitive
📌 Physics-Informed ML
└─ Applying theoretical physics frameworks to neural architecture design
└─ Symmetry-aware models, gauge theory analogues in deep learning
📌 Multi-cloud Resilience Patterns
└─ Lessons from Google & Oracle production systems
└─ Zero-downtime migration architectures
| Channel | Link |
|---|---|
| linkedin.com/in/onirudda-islam | |
| 🏢 Company | silicornya.com |
| 📅 Book a Meeting | calendly.com/cto-silicornya/30min |
| 😂 MemeMates | mememates-website.web.app |
"The limit of intelligence is not compute — it's memory."