The foundry of Abhijit Bansal

Last forged · Jul 12, 2026

I build privacy-first apps and AI agent tooling.

Designed, engineered, and shipped end to end — solo. Apps on the App Store, tools in the open, infrastructure underneath. Everything below was forged here.

See the work ↓GitHub ↗

GitHubLinkedIncontact@abhijitbansal.com

01 · Expertise

Three disciplines, one bench.

iOS & Apple platforms

Swift 6, SwiftUI, SwiftData, Vision, RoomPlan, on-device ML. Privacy-first, zero-backend architecture — apps that never phone home.

Swift 6 · SwiftUI · SwiftData · Vision · RoomPlan

AI agent tooling

Claude Code skills and plugins, MCP servers, token optimization, multi-agent orchestration. Tools that make agents cheaper, sharper, faster.

Claude Code · MCP · Rust · Python

Full-stack product craft

From Rust CLIs to design systems to App Store releases, shipped solo. The whole stack, one pair of hands.

Rust · TypeScript · Design tokens · App Store

02 · The work

Forged here.

Nine projects across apps, agent tooling, and foundation. Private builds tell their story here; public ones link to source.

Apps — iOS & macOS

Cubby

active

Inventory for garage and basement storage. NFC tags and QR labels on racks and bins — scan to see what's inside. On-device Vision suggests item names; 3D rack view, widgets, Siri. Zero backend, zero third-party dependencies.

iOS · Swift 6 · SwiftUI · SwiftDatagotcubby.com ↗

Paperix

active

Document scanner that makes searchable PDFs with on-device OCR. No cloud, no accounts, no subscription — built because PDF-scanner apps shouldn't be rent.

iOS · Swift · Vision OCRpaperix-site ↗

Floorprint

active

Scan rooms with LiDAR and RoomPlan into editable 2D floor plans. Export PDF, DXF, USDZ, GLB, STEP; a macOS mini-CAD editor for the rest. Organize the whole home in 3D.

iOS + macOS · RoomPlan · SceneKitfloorprint-site ↗

Folix

active

Privacy-first wealth dashboard: local Plaid pulls, on-device storage, AI-augmented insights. The seed of a personal financial adviser.

macOS · Swift · GRDB · PlaidPrivate build

AI / agent tooling

cartoon

active

Token-optimized CLI output for AI agents — read 12 lines instead of 800, raw logs archived. Adapters for pytest, jest, eslint, tsc; ~70% token reduction.

Rust · open sourceGitHub ↗site ↗

claude-skills

active

Skills, plugins, and agent tooling for Claude Code: iOS build loops, PM automation, prompt refinement. Installable via marketplace or standalone CLI.

Python · Shell · open sourceGitHub ↗site ↗

sift

active

Weekly AI-news pipeline: RSS ingestion, local dedup, one Claude call to curate everything into an HTML digest and a Pages archive.

Python · Claude API · open sourceGitHub ↗digest ↗

memekit

recently active

Deterministic ASCII meme reactions for CLIs, bots, and agents. 45 original formats, zero dependencies — library, CLI, and MCP server.

TypeScript · MCPPrivate build

Foundation

design-system

active

JSON tokens · CSS · Swift · Tailwind

Cross-product design tokens: one source emitting CSS custom properties, SwiftUI tokens, and Tailwind presets for every app and site in the fleet — including this one.

Private build

03 · Forge telemetry

The forge runs hot.

Everything above is built with Claude Code — and the workshop keeps its own books. Parsed locally from 3,141 session logs, May 22 – Jul 11, 2026.

305,776Lines addedacross every repo, 51 days

3,089Sessions5,656 prompts · 102,527 replies

90.4MTokens writtenoutput, straight into the melt

12.9BTokens cache-readthe crucible remembers

The works — one building per repoAll-time · May 22 – Jul 11, 2026 · 10 repos · hover a building

height = lines addedlit windows = tokens outsmoke = active this weekingots = 1 ≈ 10k linescrane = work in progresspennants = PRs merged (weekly)

01cubby8 storeys1,396 sessions · 116,975 lines · 48.0M tok out

02claude-skills5 storeys551 sessions · 49,569 lines · 11.7M tok out

03floorprint4 storeys266 sessions · 32,064 lines · 8.2M tok out

04cartoon2 storeys62 sessions · 15,617 lines · 2.9M tok out

05sift2 storeys137 sessions · 12,182 lines · 3.3M tok out

06memekit2 storeys37 sessions · 9,855 lines · 794,654 tok out

07foundry2 storeys103 sessions · 7,338 lines · 2.4M tok out

08doc-scan1 storey93 sessions · 6,042 lines · 2.8M tok out

09design-system1 storey16 sessions · 3,129 lines · 586,810 tok out

10folix0 storeys2 sessions · 0 lines · 3,019 tok out

Yard ledger: cubby — tallest hall on the yard, 116,975 lines and counting. folix — site cleared, groundbreaking soon.

Since May 1May 1 – Jul 11 · out-tokens per day

3,065Sessions

305,776Lines added

90.4MTokens out

Peak: Jun 29 — 466 sessions in one daycoolhot

Model mix102,527 assistant messages

Opus 4.8 · 38%Sonnet 5 · 26%Fable 5 · 16%Haiku 4.5 · 9%Opus 4.7 · 6%Sonnet 4.6 · 5%

Top tools

Bash22,246

Edit5,283

Write2,168

StructuredOutput1,999

30,108 thinking blocks599 subagent runs138 skill invocations/compact × 44longest prompt: 522,250 chars531 screenshots into cubby

$ memekit react --to "the token ledger"80×24

in ........... 96,285,690
out .......... 90,351,364
cache read ... 12,928,725,469

(ノ°□°)ノ ︵ ┻━┻      the ledger, first read
┬─┬ ノ( ゜-゜ノ)      the ledger, accepted

0 regrets · 305,776 lines shipped

Counted from 3,141 local session logs by a claude-skills script — on-device, like everything else here. Regenerated at each deploy.

04 · About

Solo, end to end.

I design, build, test, and ship — apps, tooling, sites. No teams, no hand-offs: the same hands that sketch the interface write the Swift, cut the release, and answer the support mail.

Privacy-first is a conviction, not a feature flag: my apps run on-device with zero telemetry.

GitHubgithub.com/abhijitbansal ↗

LinkedInlinkedin.com/in/abhijit-bansal ↗

Emailcontact@abhijitbansal.com