I'm passionate about building real, working systems — from multi-agent AI orchestration to AI infrastructure using Go, Docker, and Kubernetes. I enjoy the challenge of making complex systems reliable and well-designed.
📫 akanish327@gmail.com | LinkedIn | Portfolio
- 🎓 CSE (AI-ML) student at PES University — 8.98 CGPA
- 🤖 Building multi-agent AI systems with LangGraph and LlamaIndex
- ☁️ Upskilling in AI infrastructure & cloud-native tooling — Docker, Kubernetes, gVisor sandboxing
- 🧩 Interested in orchestration & sandboxing: Task Graph Decomposition, MCP (Model Context Protocol)
- 🏆 Rank 52, National Coding Olympiad 2021 (Codingal)
- 🌱 Currently working on Dev-Agent — a multi-agent orchestration framework with autonomous planning
| Project | Description |
|---|---|
| Dev-Agent (WIP) | Multi-agent orchestration framework with an autonomous "Plan Mode" that compiles natural language requirements into execution task graphs, using a lead-agent architecture that delegates to isolated sub-agents. Moving sandboxed execution (gVisor) onto Kubernetes. |
| MCP Servers Repository | A repository of modular MCP servers for secure, structured communication between decoupled system components — secure-by-default architecture for external tool execution. |
| Desktop AI Agent | Secure AI assistant for file management, deep research, document creation, and workflow automation via natural language, including automated end-to-end presentation generation. |
| Multi-Agent Marketing AI Agent | Specialized sub-agents coordinate to retrieve insights, update README files, publish posts, and track GitHub engagement — a fully autonomous, team-like pipeline. |
| Personal AI Agent | Autonomous agent for email management, large-document Q&A, map navigation, deep research, and web browsing — with RAG-based retrieval and local model fine-tuning via Unsloth. |
RepoLens — HSP Mentorship Program, Tilde 5.0 (Jul–Aug 2026) Selected as 1 of 4 mentees under a mentor team to build a repository-intelligence engine that turns a raw GitHub codebase into an explainable, queryable knowledge graph. Contributed to using Tree-sitter to parse source files into ASTs for precise retrieval chunks, a hierarchical batched LLM summarization pipeline (node → file → folder), Qdrant-based vector search combined with BM25 keyword search for citation-grounded answers, and an interactive dependency graph for codebase exploration.
Languages: Python, Go, C
AI/ML: LangGraph, LlamaIndex, PyTorch, Unsloth (SFT & TRL)
Infra & Cloud Native: Docker, Kubernetes, REST APIs, Git/GitHub
Orchestration & Sandboxing: MCP, Task Graph Decomposition, gVisor
Databases: MongoDB, Qdrant
- Fundamentals of MCP — Hugging Face
- RAG for Production with LangChain & LlamaIndex — Activeloop
- Machine Learning with Python — freeCodeCamp
- Python Basic — HackerRank