↑ not a screenshot. not a gif. not an iframe. one hand-written SVG file.
GitHub won't run JavaScript, so I made the picture move instead.
Rakshit Sharma — rxit. 23, Chandigarh.
I build production software alone, at a volume people usually assume is a team.
the work · the numbers · the arsenal · how i build · reach me
Live, reachable, built by me.
| SalesClawd | multi-tenant B2B SaaS — autonomous agents acting on real marketing systems |
| NY Elizabeth | live bidding platform for a Beverly Hills auction house |
| Aumiqx | AI agents that actually do the work |
| Rockport India | footwear ecommerce |
| air up | German DTC brand |
| Beast | video production studio, London |
| Yoginii | women's wellness marketplace |
131,348 lines of code · 796 commits · 46 days committed · 2,855 lines per active day · 92% TypeScript · 44,250 largest single build · 52.4% of commits after 22:00
Forty-six days, spread across a hundred and fifty-three. I don't grind a streak — I disappear, then put down about 2,855 lines in a day.
Computed by scripts/measure.mjs in this repo, across 22 repositories, counting only code — markdown, JSON and config excluded. Not typed by hand. Clone it and check.
salesclawd — 44,250 lines · autonomous agents acting on real marketing systems
The problem. Agencies promise AI and ship dashboards. Something that reports is not something that works.
What I built. Multi-tenant B2B SaaS where the agents hold real credentials and actually act on a client's live marketing systems.
The hard part. Tenant isolation when your agents have write access to production. A leak here isn't a bug report, it's someone else's customer data.
Result. My largest single build — 44,250 lines across 228 commits. Live.
bhraman — 20,362 lines · travel product
The problem. Planning a trip means ten tabs and a spreadsheet you abandon by day three.
What I built. One surface for planning and booking.
The hard part. Modelling itineraries that survive contact with reality — flights move, places close, plans change at 2am in an airport.
Result. 20,362 lines across 44 commits.
assay — 12,023 lines · data governance, PII classification, quality scoring
The problem. Nobody actually knows what's inside their own datasets. "Is there PII in this?" is answered by vibes.
What I built. Ingest any dataset; auto-discover its structure, flag sensitive fields, then score it for quality, trust and value.
The hard part. Classifying PII without the data ever leaving the box. The moment you solve governance by shipping the data somewhere, you've become the problem.
Result. 12,023 lines across 59 commits.
orbit — 7,245 lines · local-first personal CRM and life-OS
The problem. Every CRM wants your life on their servers, forever, at $29/month.
What I built. A local-first CRM and task system that runs entirely on my own machine.
The hard part. Local-first sync with no backend to arbitrate conflicts. You don't get to punt on the hard case.
Result. 7,245 lines. I use it every day, which is the only review that counts.
sift — 2,597 lines · semantic search and cited chat over your own library
The problem. Saved articles are where reading goes to die.
What I built. Save any URL or note, then search it and ask questions — with citations.
The hard part. Answers that cite their source, and refuse when they can't. A RAG system that would rather say nothing than make something up.
Result. 2,597 lines.
sightline — 2,210 lines · auditable, approval-gated media desk
The problem. Affiliate spend moves faster than any human can verify it.
What I built. A tracking-integrity auditor and cross-platform P&L, with approval gates in front of anything that moves money.
The hard part. Making every automated action reviewable before it spends, not after. Automation you can't stop isn't automation, it's exposure.
Result. 2,210 lines across 28 commits.
Also on the shelf: traviqx · openbdr · job-radar · remote-hunt · skillmap · tellur · beacon · forge
flowchart LR
P([problem]) --> S[spec<br/>written first]
S --> PL[plan<br/>broken into tasks]
PL --> F{fan out}
F --> A1[agent]
F --> A2[agent]
F --> A3[agent]
A1 --> V[verify<br/>one runnable check]
A2 --> V
A3 --> V
V -->|fails| PL
V -->|passes| SH([shipped])
The expensive model plans. Cheap models execute in parallel. Nothing merges without a check that fails when the logic breaks.
Everything else is taste: local-first before SaaS, free tier before invoice, the standard library before a dependency, and deleting code before writing more of it. This README has zero runtime dependencies for the same reason.
What I actually reach for — TypeScript, Node, Fastify. Postgres with row-level security, Drizzle or Prisma. BullMQ and Redis when work needs a queue. Next.js and React on the front. Python when the problem is shaped like Python. The Claude API with plan/execute cost routing, MCP, and multi-agent orchestration.
when I actually work
52.4% of my commits land between 22:00 and 06:00. Peak hour is 01:00, at 86 commits. My single busiest day was 18 May 2026: 96 commits.
I'm aware of what this looks like. I'm not defending it. It's just what the data says, and I said every number here would be real.
the graveyard
openclaw — died 17 July 2026.
The whole idea rested on a single assumption. The assumption turned out to be false, so the project went, and every reference to it was purged the same day.
Shipping is a skill. Killing things is the same skill pointed the other way, and it's the one nobody puts on a portfolio.
the honest stack opinions
- Postgres row-level security over app-layer authorization. The database is the only place a rule can't be forgotten by a new endpoint.
- Local-first over another subscription. If it works on a plane, it works.
- OpenStreetMap over a paid lead database. The data is the same data.
- The standard library over a dependency, and one line over fifty. Every package is a promise someone else has to keep.
- Delete first. The best code is the code that never had to exist. Half of building well is refusing to build.
If you're building something genuinely hard, I'd like to hear about it.
rakshitsharma4179@gmail.com · rxits@proton.me · github.com/rxits
