The Scale of Operations
At any given time, 11+ active projects are running across 4 domains. That is not a workload a 19-year-old CS student handles alone without the right tools. Claude Code is how it's possible.
| Domain | Projects | What Claude Code Does |
|---|---|---|
| Civic Tech | Open Permit | Codebase dev, pitch decks, funder outreach, legal frameworks, data pipelines |
| Healthcare AI | Knight Medicare | Full stack dev, Supabase schema, abstraction layers, clinical data modeling |
| EdTech | Adventurers Guild | Next.js/Prisma codebase, financial modeling, feature specs, PM audit |
| Open Source | Open Paws repos | RAG systems, data pipelines, platform infra |
| OSS Contributions | external + Open Paws repos | Codebase understanding, PR preparation, test writing, review responses |
| Infrastructure | Cross-project | Yotta migration, Supabase alternatives, self-hosting analysis |
| Fundraising | Open Permit + broader | Grant pitches, funder databases, outreach campaigns |
| Career | Portfolio, applications | Resume, portfolio updates, program applications |
5 Patterns
Architecture Sessions
When starting a new feature: open a session with full context — what exists, what needs to exist, what could go wrong. The back-and-forth surfaces edge cases and challenges assumptions before a line of code is written.
Data Pipeline Development
A surprising amount of work is data pipelines, not UIs. 197-contact funder CSVs for Open Permit, financial modeling for the Guild. All managed through Claude Code sessions.
Pitch & Strategy Work
AARC Demo Day decks, SFF grant framing, Dasra outreach strategy, Adventurers Guild financial model. Non-technical work is a large fraction of total Claude Code usage.
OSS Contribution Workflow
Codebase understanding, PR description drafting, test convention matching, review iteration. This workflow supports contributions across scientific Python, sustainability tooling, developer infrastructure, and Open Paws production repos.
Infrastructure Decisions
The Yotta evaluation (29 services, 11 projects, 63% savings), SMS provider selection, Supabase alternatives analysis — complex tradeoff evaluations managed collaboratively.
The Meta-Pattern: Institutional Memory
"Across 100+ conversations, Claude Code has accumulated context about how all 11 projects relate. Each new session starts with context. That compounding context is the real value — not any individual code generation, but the coherent understanding of how Knight Medicare's safety requirements inform Open Permit's data handling, how the Guild's rank system maps to Open Paws' intern pipeline, how the Yotta migration affects all of them."
Without that compounding context, running 11 projects means constantly re-explaining architecture decisions, re-justifying product choices, and re-litigating infrastructure tradeoffs. With it, each session builds on the last. The velocity compounds.
What Claude Code Doesn't Do
- Clinical decisions — Knight Medicare's clinical protocol is the Harvard PhD's domain. Claude Code doesn't touch disorder classification or therapeutic approach selection.
- Relationship building — Funder meetings, hackathon networking, team management, and mentor relationships require human presence and judgment.
- Creative vision — "What should we build next" decisions come from product intuition and domain knowledge, not AI sessions.
- Shipping discipline — Deciding what to ship today vs. next month, what to cut, what to delay. Prioritization is a human judgment call.
- Accountability — When code breaks at 2 AM, when a deadline slips, when a grant application gets rejected — that's on me. Claude Code accelerates the work; it doesn't share the consequences.
The Workflow in Practice
Morning Block (2-3 sessions) ├── Architecture session for current sprint priority ├── Code review / PR iteration on OSS contributions └── Data pipeline work or CSV processing Afternoon Block (2-3 sessions) ├── Feature development (main project in sprint) ├── Pitch/strategy work or funder outreach material └── Cross-project infrastructure decisions Evening Block (1-2 sessions) ├── OSS PR preparation for new contributions ├── Documentation and planning for next day └── Review feedback / iterate on open PRs Weekly Cadence ├── Architecture review session per major project ├── Financial model update (Guild/Open Permit) ├── Open Paws operations and India entity work └── Portfolio and career materials update
By The Numbers
Projects Under Active Development ├── Civic Tech: Open Permit (8+ countries, 200+ permits) ├── Healthcare AI: Knight Medicare (27 disorders, 10 modalities) ├── EdTech: Adventurers Guild (500+ users) ├── Open Source: Open Paws (RAG, LEWS, infra) └── Cross-cutting: Infrastructure, fundraising, career Sessions Per Day: 6-7 average Force Multiplier: 4-5x (conservative estimate) OSS PRs Supported: reviewer-ready PRs across multiple repos Pitch Materials: AARC Demo Day, SFF grant, 197-contact funder DB Infrastructure Work: 29 services, 11 projects, Yotta migration Non-code Sessions: ~40% of total usage
Key Lesson
Claude Code isn't a replacement for engineering skill — it's a multiplier. Batman doesn't fight crime because he has gadgets. He has the judgment to use them effectively, and the gadgets extend his reach. Claude Code is my Batcave. The judgment about what to build, how to prioritize, what tradeoffs to make — that's still entirely human. The execution velocity is not.