Interest Map
This map summarizes recurring technical interests visible in SWAP and the current AI/ notes.
AI Agent Runtimes and Orchestration
You repeatedly explore how agent systems should be structured once they move beyond one-off chat. The strongest thread is an orchestrator-first model: a lean main session for intent, decisions, and review, with workers handling heavy execution. This appears in Pathfinder Orchestration Architecture and is operationalized further in Bastion - Cursor as Primary Runtime Plan.
Related concepts:
- agent orchestration
- worker delegation
- ACP
- Cursor-backed execution
- runtime migration
- main-chat control plane
Personal Agent Memory and Curiosity Systems
The clearest self-referential R&D thread is the idea of a personal curiosity system that mines notes, tracks open loops, ranks topics, and composes high-signal prompts. Pathfinder - Proposed Curiosity Agent Stack explicitly describes the desired artifacts: interest-map.md, open-loops.md, curiosity-backlog.md, and recently-explored.md. Pathfinder - Curiosity Stack Implementation turns that into a more concrete state model and agent contract design.
Related concepts:
- second-brain maintenance
- curiosity backlog
- topic ranking
- anti-slop filtering
- personal R&D frontier
Agent Observability and Evaluation
Several notes point toward treating agent work as something that should be traced, scored, and improved. OpenClaw Observer VM for Agent Observability frames this as async review, evaluator loops, event schemas, scorecards, and postmortems for orchestrator and worker behavior. This intersects with the R&D frontier because it would let the system learn from actual agent runs rather than only from static notes.
Related concepts:
- LLM observability
- evaluator/judge loops
- async postmortems
- trajectory analysis
- workflow optimization
- Tailscale-reachable observer VM
Observability Data Infrastructure
ClickHouse MCP for Dataservice Trace Analysis explores ClickHouse, OpenTelemetry, and MCP as the basis for trace investigation. The interesting pattern is not raw database chat, but a curated semantic layer over observability data: views, saved queries, parameterized tools, and trace-specific helper functions.
Related concepts:
- ClickHouse
- OpenTelemetry
- MCP
- trace analysis
- incident copilots
- curated semantic tools
Software Systems and Backend Depth
SWAP contains a cluster of backend and infrastructure topics: Kafka, Elasticsearch, OLAP vs OLTP, VPCs and Kubernetes, sidecars, Spring Boot vs Spring Cloud, Tomcat chains, JVM/GC, SSO, SMS vs APN, and Postgres/serverless coupling. These are not isolated facts; they point toward a durable interest in understanding the operational shape of production systems.
Related concepts:
- distributed systems
- data infrastructure
- backend frameworks
- cloud networking
- storage engines
- system design interview depth
Personal Data Products and Finance Automation
SWAP now contains a detailed project seed for personal spending ingestion and analytics. It connects finance, data ingestion, reconciliation, Apple platform constraints, email parsing, bank statements, rewards optimization, and frontend analytics. This is one of the most concrete buildable project ideas in the current source set.
Related concepts:
- personal finance analytics
- ingestion pipelines
- reconciliation algorithms
- Apple Shortcuts
- email parsing
- card reward caps
Business, Markets, and Institutions
SWAP also has repeated questions around information, companies, IPOs, stock exchanges, credit cards, Meiji Restoration, and why successful companies succeed. This is less directly connected to the AI notes, but it suggests a parallel curiosity track around how institutions, incentives, and markets shape technology.
Related concepts:
- business strategy
- market structure
- finance systems
- company history
- information value