Bridging high-precision financial logic with autonomous AI systems
I architect agentic AI workflows and scalable backend systems — transitioning from a career in Financial Trading and Data Engineering (PySpark, SQL, Airflow) into the AI automation space.
My edge: I bring financial-grade precision and data-pipeline thinking to every system I build. I don't copy-paste — I go deep on fundamentals (Linux internals, API design, SOLID principles, MCP security) so I can architect, build, and ship independently.
"Done is better than perfect, but clean code is non-negotiable."
AI & Automation
Backend
Data Engineering (Heritage Stack)
Infrastructure & DevOps
| Area | What I'm Building Toward |
|---|---|
| 🐹 Go (Golang) | High-performance backend services — concurrency model, stdlib depth, production-grade CLI tools |
| ☸️ Kubernetes | Graduating AI agent workloads from local Docker to resilient, auto-scaling K8s clusters |
| 🤖 MCP & Agentic Systems | Custom MCP servers that bridge LLMs with local tools and external APIs securely |
| 🐧 System Fundamentals | Linux internals, Bash automation, and OS-level performance tuning on Ubuntu |
Go · Neo4j · PostgreSQL + pgvector · SQLite · MCP · Docker
A high-performance long-term memory hub for AI agents. Shard-Link acts as a semantic gatekeeper — fragmenting, storing, and routing context across a Triple-Engine architecture (graph mesh + vector archive + local seed memory) so LLMs maintain persistent, privacy-first memory across sessions.
Key engineering highlights:
- Knowledge Mesh (Neo4j + GDS): Graph-based relational reasoning with PageRank centrality and Louvain community clustering for multi-resolution retrieval
- GraphRAG Synthesizer: Background goroutine autonomously bonds resonant shards, detects communities, and generates LLM-summarized macro-context — without blocking the MCP server
- Survival Formula: Every shard gets a retention score
S = (Density × Centrality × Vitality) / TimeDecay— the system reinforces what you think about most and evicts orphaned data automatically - Sub-5ms retrieval target: SIMD-accelerated vector search via
pgvector, database-side computation, minimal Go overhead - Defense-in-depth security: Cloudflare Tunnel edge → API key middleware → HTTPS transport. No port forwarding required
- MCP Streamable HTTP (2024-11-05): Primary transport with 10s heartbeat intervals to survive Cloudflare idle timeouts; SSE kept for legacy compat
📄 Whitepaper · Roadmap · Repo →
n8n · Node.js · MCP
Custom MCP servers that give LLMs secure, structured access to local tools and external APIs. Focus on autonomous task execution with proper error boundaries and least-privilege API integration.
Next.js · NestJS · Clean Architecture
Netflix-style movie platform built as a full-stack monorepo. Demonstrates domain-driven design, type-safe backend contracts, and clean separation of concerns across the frontend/backend boundary.
I'm open to roles and collaborations at the intersection of AI Automation, Backend Engineering, and Cloud-Native infrastructure.