Writing
Notes on the systems Owais Barkati builds — streaming data, change data capture, multi-agent orchestration and the parts of applied machine learning that only show up in production.
- 01
Debezium to ClickHouse: sub-second CDC without losing consistency
Change data capture delivers at-least-once, and an analytical store counts duplicates rather than correcting them. Here is why the binlog beats a batch extract, where deduplication actually belongs, and the two failure modes — schema evolution and silent snapshot gaps — that break pipelines quietly.
- CDC
- Debezium
- ClickHouse
- Kafka
- data-engineering
- 02
Federated fine-tuning of Llama 3 with Flower and QLoRA
Federated learning moves the model to the data instead of the data to the model. That inversion is only practical for large language models because QLoRA shrinks both what each client must hold in memory and what it must send back. Here is why the two techniques belong together.
- federated-learning
- QLoRA
- Flower
- LLM
- fine-tuning
- privacy
- 03
LangGraph supervisor routing: getting a query to the right agent
One agent holding every tool degrades as the toolset grows. A supervisor that classifies intent first and dispatches to a specialist keeps each agent's context narrow and each failure attributable — but routing becomes the single point where the system can be wrong beyond recovery.
- LangGraph
- multi-agent
- LLM
- routing
- FastAPI