About this curriculum
Flow Research is evolving. Our research teams explore distributed systems, AI, and protocol infrastructure, and the work feeds into products still taking shape. This curriculum is the learning layer of that ecosystem — a structured path for engineers who want to build the kind of public-good technology these products depend on.
Things are moving fast, and this will change as we do.
Products
The team is building a set of products that work as one system. Here is how they fit together:
- Jarvis — the agent runtime. Spawns, configures, and secures Personal Operators so they can connect to the Flow Research economy.
- Garden — the human-agent workspace. A persistent space where people and agents collaborate with connected tools, workflows, and approvals.
- WorkStream — the task pipeline. Takes work from economic value sources, distributes it to humans and agents, verifies outputs, and handles attribution and rewards.
- Harnessy — the reliability layer. Tests agent behavior, evaluates task output, and closes the feedback loop so agents can be trusted with real work. Jarvis gives the agent life. Garden gives the agent a workspace. WorkStream gives the agent and human valuable work. Harnessy makes the agent reliable.
Contributing
Flow Research runs a fellowship program for engineers who want to help build these products or explore new ideas through the research track. Contributions are tracked in FlowLedger — points, badges, and a public ledger that makes every contribution visible and rewardable.
For now, the curriculum covers these areas:
Foundations
How to learn, take rigorous notes, write design docs, and build a public portfolio.
- Concepts — learning hierarchy, maieutic thinking, student vs engineer mindset
- Practice — reading comprehension, effective notes, building a portfolio
- Tooling — markdown, version control, collaboration workflows
- Specification — design docs, ADRs, API specs, research notes
Start here if you're new.
Blockchain
From fundamentals through smart contracts and security to protocol engineering and scalability.
Beginner — what is blockchain, layer 1 vs layer 2, tokens and incentives, decentralized identity
Intermediate — smart contract design patterns, testing and deployment, common vulnerabilities, code audits, pen-test workflows
Advanced — protocol architecture, consensus tuning, governance mechanisms, state channels, rollups and sharding, interoperability
AI/ML
Understand, use, and govern AI-driven components in production systems.
Beginner — math for ML, data pipelines, model lifecycle, Python ecosystem, notebooks, ML libraries
Intermediate — supervised learning, feature engineering, hyperparameter tuning, CI/CD for models, monitoring and drift, deployment patterns
Advanced — transformers, graph neural networks, reinforcement learning, paper replication, model alignment, ethics and responsibility
Protocol Engineering
Design, implement, and evolve reusable protocols that systems communicate over.
Beginner — protocol vs application, state machines, communication patterns, specification writing, versioning, interoperability
Intermediate — compatibility testing, scaling design, resilience patterns, upgrade paths, interchain protocols, community feedback
Advanced — latency optimization, consensus economics, security modeling, regulatory compliance, performance auditing, enterprise integration
Rust Engineering
Design, ship, and operate reliable Rust systems.
- Rust mindset, toolchain, and engineering loop — compiler feedback, project loop, and a first CLI artifact
- Ownership, borrowing, lifetimes, and memory thinking — ownership as system design
- Data modeling, errors, and control flow — structs, enums, state machines, and typed errors
- Reuse without OOP — traits, generics, newtypes, and composition
- Axum-first web engineering — typed handlers, extractors, state, responses, and middleware
- Persistence and reusable CRUD with SQLx — explicit SQL, repositories, services, transactions, pagination, and scoped resources
- Production service capstone — build and review a portfolio-grade Rust service
This path is general-purpose: it is useful for backend services, CLIs, workers, protocol tooling, data systems, and infrastructure projects.
Agent Systems
How agents work under the hood — orchestration, tool use, memory, planning, safety, and evaluation.
- What are agent systems? — core components and how they fit Flow Research
- LLM orchestration — prompts, chains, routers, tool loops
- Tool calling and integration — tool design, registry, security
- Memory and state — short-term, long-term, episodic, procedural
- Planning and reasoning — approaches, failure recovery, reflection
- Safety and guardrails — input/output guardrails, escalation, kill switches
- Evaluating agents — correctness, faithfulness, adversarial testing, human eval
This area maps directly to Jarvis, Garden, WorkStream, and Harnessy — agents are the core of every Flow Research product.
How to use this
If you're just starting out — begin with Foundations, then go deep in whichever area fits your interest.
If you're experienced but new to this stack — skip what you know, fill the gaps.
If you're building a real project — map your project across the curriculum. After each lesson, refactor your spec or diagrams.
Use the exercises at the end of each lesson to build a portfolio you can show others.
What this assumes
You are willing to write, diagram, and code as you go. You care about maintainability and understandability, not just making things work. You do not need a PhD or a fancy title — you do need the habit of shipping small, explicit artifacts that show what you've learned.
Next steps
Start with Foundations, or jump straight into an area: