/ Think. Deisgn. Code. Deploy.
AI, Embedded, Blockchain — Shipped, Not Theorized.
Diop Vector lab is a living record of rmy current projects : architecture decisions, integration failures, and knowledge base code across every domain I am interested in. The moto is Signal over noise — always.
Processes currently running
LLM Engineering
ESP32 + FreeRTOS
Solidity + Bitcoin
RAG + Agent Systems
PyTorch, LangChain, PydanticAI — fine-tuned pipelines with documented failure modes, not benchmark theatre.
Bare-metal firmware, MQTT telemetry, and constrained inference — documented at the hardware limit.
Smart contracts, on-chain state machines, and protocol-level analysis — not wallet tutorials.
Retrieval pipelines, vector databases, and autonomous agent loops — traced end-to-end with real data.
Kafka Streaming
Docker + Kubernetes
Grafana + FastAPI
Sensor + Edge Systems
Event-driven pipelines at scale — partitioning strategy, consumer lag, and the production failures that clarified the design.
Container orchestration, GitOps workflows, and cluster-level observability — reproducible from manifests.
Metrics pipelines, live dashboards, and API instrumentation — built to surface what matters under load.
Hardware-to-cloud telemetry pipelines — firmware, protocol, and ingestion layer designed as one system.
• Golang
Systems-Grade Python
Low-Level Systems
Distributed Infrastructure
Async patterns, typed interfaces, and performance-critical Python — where the language is a constraint, not a convenience.
Memory-safe systems code for embedded targets and high-throughput services — ownership model as a design tool.
Multi-region deployments, service mesh, and cost-aware architecture — decisions documented at each scale boundary.
Code that runs in production
On-Device Inference Engine
Real-Time Event Pipeline
On-Chain State Machine
TensorFlow Lite model quantized to run on ESP32 under FreeRTOS. Three quantization strategies failed before the fourth held. Constraints documented.
Kafka-backed streaming architecture processing sensor data at sub-100ms latency. Partition rebalancing failures shaped the final topology.
Solidity smart contract implementing a verifiable state machine with gas-optimized transitions. Two audit findings drove the architecture pivot.
Syntax-highlighted, searchable, and documented where it fails — every entry covers a real integration problem across ESP32, Kafka, LangChain, Solidity, Kubernetes, and more.