01
A sync architecture built to recover
Developed and maintain a distributed producer-consumer system with high visibility and self-healing state management.
10k messages/minProduction syncing with observable, self-healing state
Backend · distributed systems · production inference · applied ML
Make
Break
Rinse and repeat.
I occasionally dabble in vibe-coded web.
Currently building @ Grexa AI.
Selected systems · Grexa AI
Case files.
Systems are mostly invisible. Scale, failure modes, and engineering decisions are not.
02
Agent automation across 5,000+ locations
Designed the backend for the Grexa GBP Agent that handles reviews, posts, schedules, and other ongoing profile optimization tasks on a client's Google Business Profile.
5,000+ locationsScheduled operations and automatic review replies
03
Is it water, rewritten in Go
Reimplemented a third-party TypeScript API (Is it water) to remove dependency constraints and make an internal integration faster and easier to own.
Lower latencyFewer external constraints; clearer internal ownership
Working set
Tools I reach for.
The compact core of my production stack: a language, document store, cache, and message broker.
Go
MongoDB
Redis
RabbitMQ
Research & applied ML
From models
to evidence.
I
Volatility based reinforcement learning
Published in European Economic Letters, Vol. 15, Issue 3 (2025). The proposed high-frequency trading strategy outperformed buy-and-hold on 414 of 750 stocks.
II
Evolutionary-ensemble based AQI prediction
Presented at SOCTA 2024, published in Springer Nature. Genetic-algorithm-optimized ensemble learning achieved results comparable to deep LSTM and GRU models, with R2 = 0.97.
III
Provider integrations, prompt and context management, knowledge bases, RAG pipelines, LLM visibility.