Hi, I'm Nasit Sony 👋
💡 AI Infrastructure Engineer | Distributed Systems | Consensus Protocols (BFT)
📍 Building production-grade backend systems for AI and fault-tolerant distributed computing
I build distributed backend systems where correctness matters — even under failures, partitions, and adversarial conditions.
My work sits at the intersection of:
- AI infrastructure
- Distributed systems engineering
- Byzantine fault-tolerant consensus
I focus on treating AI systems not as simple APIs, but as fault-tolerant distributed systems with strong correctness guarantees.
💰 Production Systems (Fintech)
- Built international money transfer systems handling $600M+ annual volume
- Focus: correctness, consistency, and performance under real-world constraints
🔬 Distributed Systems & BFT Research
- Published work in Springer journals and international conferences
- Designed and implemented Byzantine fault-tolerant protocols
- Bridging theoretical guarantees with practical system design
🚀 Current Focus — AI Infrastructure
- Building production-grade AI pipelines (SmartSearch)
- Applying distributed systems thinking to RAG and ML systems
- Exploring decentralized and fault-tolerant ML infrastructure
Production-oriented AI system built like real backend infrastructure.
Key Features:
- Asynchronous ingestion (Kafka → Workers → PostgreSQL)
- Embedding-based semantic search (pgvector + OpenAI)
- Retrieval-Augmented Generation (RAG)
Reliability Engineering:
- Idempotent processing (no duplicate chunks)
- Retry + DLQ handling
- Worker crash recovery (offset-safe Kafka processing)
- Explicit lifecycle tracking (PENDING → PROCESSING → READY/FAILED)
- Eventual consistency guarantees
👉 Focus: AI systems as distributed systems
Backend infrastructure system focused on security, scalability, and reliability.
- JWT & OAuth2 authentication
- Request routing and service orchestration
- Observability and failure handling
Storage engine inspired by LSM-tree designs (RocksDB-style).
👉 Focus: durability, compaction, and performance trade-offs
Fault-tolerant distributed system inspired by Raft and BFT protocols.
👉 Focus: safety, liveness, and failure handling
Languages:
Java, Rust (learning), Python
Backend & Infra:
Spring Boot, Kafka, PostgreSQL, pgvector
AI Stack:
Embeddings, RAG pipelines, OpenAI APIs
Distributed Systems:
Consensus, fault tolerance, idempotency, retries
- Prioritized-MVBA — Optimal Asynchronous Byzantine Agreement Protocol
- Published in Springer journals & international conferences
- Google Scholar: https://scholar.google.com/citations?user=mBIQ1-0AAAAJ&hl=en
- Medium: https://medium.com/@nasitsony96
- Topics:
- Failure handling in async systems
- Idempotency and retries
- Designing production-grade RAG systems
- AI infrastructure & reliability
- Fault-tolerant distributed systems
- Consensus protocol engineering
- LinkedIn: https://www.linkedin.com/in/nasitsony
⭐ I believe systems should be designed for failure — not just success.