Johan Hellman · Stockholm / Madrid
Accordant is a personal workshop exploring what happens when AI is taken seriously — not as a coding assistant bolted onto an existing workflow, but as the primary builder, operator, and analytical engine for real problems in finance and tax.
Try Rán — Spanish IRPF explainer →What I'm building
Spain's annual tax return (Modelo 100) is four pages of AEAT jargon that most filers sign without understanding. Rán takes the draft PDF and explains everything in plain language: what every figure means, how the calculation works, what's worth a second look. Built for expat professionals and international families navigating cross-border complexity.
A deterministic portfolio rebalancing engine that started as an offline calculation core and evolved — through AI-driven development — into a fully autonomous live agent. Streams portfolio drift in real time, proposes trades, and executes via Alpaca with circuit breaker safety limits and immutable JSONL audit trails.
AGENTS.md ruleset
Go command-line tools for the Alpaca API: broker operations and retail/paper trading. Built to support the rebalancing engine. Installable via Homebrew.
OpenClaw · Agent team
All projects share an AI-native infrastructure built around OpenClaw. Rather than one general-purpose assistant, the setup runs a team of specialised agents — each with a defined scope, trust level, and operating model. Agents work autonomously within their brief; anything consequential requires explicit approval.
Executive assistant and main coordinator. Manages tasks, routes work to specialists, and maintains continuity across sessions.
Daily ingestion of investment signals from curated financial accounts on X. Scores, deduplicates, and routes signals for review.
Processes bookmarks and research inputs. Classifies, routes, and drafts decision records when external content warrants a response.
Portfolio analysis, signal synthesis, and thesis validation. Consumes Heimdal's signals, feeds the rebalancing engine. Execution requires explicit approval.
Reviews plans, specs, and product documents for technical soundness and strategic alignment. Read-only — no external actions.
Names follow Norse mythology — chosen to point toward what each agent should become, not just what it does today.
Philosophy
Each project treats AI as a first-class collaborator, not a productivity tool. The rebalancing engine's architecture, tests, and documentation were all produced through structured LLM collaboration. Rán uses Claude as its analytical engine. The OpenClaw harness runs agents that monitor markets, route research, and review code autonomously.
The consistent pattern: deterministic, auditable logic at the core, with AI as the reasoning layer above it. Not AI generating code that humans maintain — AI as an ongoing operator of systems that remain legible and verifiable.
Finance and tax are the starting point because the rules are codified, the stakes are concrete, and the gap between what professionals have access to and what individuals have access to is large. You know immediately whether the output is any good.