Live code. Real constraints. No mockups.
Every project here links to a running repo or deployed endpoint. Architecture diagrams show integration across AI, embedded hardware, blockchain, and distributed infrastructure — not what was planned, but what shipped.
All · AI & ML · Embedded Systems · Blockchain · Distributed Systems · DevOps & Infrastructure · IoT · Real-Time Streaming
TinyML Inference on ESP32
Real-Time Kafka Telemetry Pipeline
On-Chain Audit Trail for IoT Sensor Data
Quantized neural network running on-device with FreeRTOS task scheduling. Memory ceiling forced a novel weight-pruning pass before deployment.
Sub-100ms end-to-end latency across 12 partitions. Three architectural rewrites exposed where the consumer lag actually originated.
Solidity contracts verify sensor hashes before writing to chain. Gas optimization required rethinking the commit-reveal scheme entirely.
ESP32 · FreeRTOS · TensorFlow Lite · MQTT · C++
Kafka · Python · Grafana · Docker · Kubernetes
Solidity · Ethereum · MQTT · Rust · Hardhat
Autonomous LLM Agent with Tool Use
GitOps-Driven Kubernetes Cluster
Production RAG Pipeline with Hybrid Retrieval
LangChain orchestration layer driving external API calls and vector-database lookups. Failure modes catalogued across 400 production runs.
Full declarative cluster state managed via Flux CD. Incident post-mortems from three failed rollbacks drove the final reconciliation strategy.
Dense and sparse retrieval combined over a 2M-document corpus. Reranking latency was the bottleneck — not embedding, not generation.
LangChain · PydanticAI · FastAPI · Qdrant · Python
Kubernetes · Flux CD · Terraform · Prometheus · Helm
PyTorch · Qdrant · FastAPI · LangChain · Docker
Every repo includes the readme that documents what broke first.
Constraints drove every decision documented here. The source is open — architecture notes, failure logs, and reproducible build instructions are in the repo alongside the code.