Architecture audit + Claude Code workflow review for production SaaS on Next.js / tRPC / Prisma

Hello, Taras!

Your stack — Next.js, React, Prisma ORM, TypeScript, PostgreSQL — is exactly what I work with every day. In my portfolio, there is a project Ukrainian Cuisine built from scratch to deployment on Vercel.

What I will do as part of the audit:

Findings register: I will document all findings in F-NN format with severity, effort, file refs, root cause, and a ready Claude Code prompt for corrections.

ADRs: I will describe non-obvious architectural decisions in your ADR-NNN format with continued numbering.
…
CLAUDE.md updates: specific rules and anti-patterns tailored for your codebase.

Refined 4–8 week plan: in NOW.md / ROADMAP.md format with a breakdown by hours and revenue terms.

Claude Code workflow notes: where Claude Code generates tech debt in your repo and how to fix it through slash commands and hooks.

Bonus: 1–2 SPEC-NN handoff docs for the next modules.

I use Claude Code every day — I understand the context from the inside.

Show original

Translate

Hello!

This is a very interesting request, especially because you need not a generic audit, but an audit that will actually fit into the existing documentation/code workflow and be useful for Claude Code after the review.

I am a full-stack JavaScript/TypeScript developer with experience in React / Next.js / Node.js, Prisma, SaaS projects, AI workflows, and production codebases. I also actively work with an AI-assisted development approach: specs → implementation → verification → docs/handoff, where AI does not just "generate code," but operates within clear rules, context, and documentation.

One relevant project: PrecisAI — a full-stack AI/SaaS platform on Next.js, Supabase/Postgres, Vercel, and several LLM APIs. The most challenging part was not the AI integration itself, but controlling context flow, agentic workflow, media monitoring pipeline, debugging AI outputs, and reducing tech debt when many decisions are quickly generated by AI. I would have formalized specs, ADRs, agent rules, and verification checklists earlier so that AI does not create hidden architectural inconsistencies.

My actual AI/Claude-style loop:
…
1. First, I read the code and documentation, not starting with generation.
2. I form a short intent/spec: what we change, what we do not touch, what constraints exist.
3. I give AI a narrow execution scope, not "fix everything."
4. After execution, I check the diff, types, tests, edge cases, API contracts.
5. If the solution is non-obvious — I document it in ADR/spec/doc, so that subsequent iterations do not break context.
6. For large changes — I break them into small handoff prompts that can be run separately.

What I would question in your announcement: the requirement that "everything must be immediately suitable for Claude Code without rewriting." I agree with the goal, but in practice, part of the documentation needs to be normalized first: removing ambiguity, duplicates, outdated rules, and conflicting instructions. Otherwise, Claude Code will use the documentation as a compass, but a compass can point in different directions.

What I can prepare:

* findings.md in F-NN format with severity, effort, path:line, root cause, fix proposal, and Claude Code handoff prompt;
* ADR-NNN for lock-in / reverse decisions;
* targeted updates for CLAUDE.md specifically for your codebase;
* a 4–8 week plan in NOW.md / ROADMAP.md format;
* workflow notes: where Claude Code creates debt, what prompts/hooks/subagents/skills should be added;
* 1–2 SPEC-NN handoff docs for future modules.

I will not propose rewriting the stack. My focus is to understand the existing system, identify weak points, document solutions, and ensure that Claude Code works more accurately after the audit.

Show original

Translate

Hello. I am interested in the project: a solo product on Next.js/tRPC/Prisma that has been in production for 6 months with Claude Code as the main engineering force. This is a format that requires real engineering empathy, not just a template audit.

I have 3.5 years of commercial full-stack experience in TypeScript, Next.js/Nuxt, NestJS, PostgreSQL. I have worked on a B2B e-commerce platform (~40 models, complex domain logic) and products at the intersection of UX and architecture. I actively use Claude Code every day: I build the structure of documentation, outline invariants, and navigation through context so that the model works effectively within the codebase rather than rewriting it.

I can conduct an audit in a format that fits into the existing documentation system, with findings, recommendations, and concrete steps suitable for Claude Code as a compass. I am ready to discuss the scope and priorities in detail.

Show original

Translate

You have filled the text with English words so much that you could have already submitted the application in English. Code audit, no questions. Write to us, we will evaluate and do it.

Show original

Translate

Hello, Taras.

1. Solo: cyanship.com, SaaS solo on a close stack. Auth, Stripe billing, streaming AI chat, usage-based balance. Complicated: synchronization of Stripe webhook with internal balance under race conditions. I would currently move AI quotas from Mongo to Redis with atomic decrement instead of optimistic transactional update.

2. Claude Code loop:
– planning: task plus CLAUDE.md context, I request 3 approaches with trade-offs, I choose one.
– spec: input, output, edge cases, acceptance in a separate .md, I fix it to the code.
– execution: atomic commits of 30-90 minutes, after each types plus relevant tests.
– verification: if the result does not match the spec, I correct the spec, I do not apply patches on top.
…
3. Push back on bonus: 1-2 SPEC-NN docs within the audit give bias. First, prioritization of findings and decisions fix/refactor/leave, after that specs are written for the most important, not for the most convenient. I would separate the bonus as a separate stage after agreeing on the register.

Stack: Next.js, Prisma, Redis, TS daily. tRPC, Hono, Bun, BullMQ, Better-Auth, CASL not daily, I have worked on close analogs, I read code. I say it as it is.

Give read-only access to 2-3 most problematic files, I will return specific notes within 24 hours.

Stages:

1. Read-through of the codebase, initial findings, agreement on priorities — 12,000 UAH
2. Findings register, ADRs, CLAUDE.md updates — 12,000 UAH
3. Refined 4-8 week plan, handoff documents — 6,000 UAH

Show original

Translate

Hello! I have carefully reviewed your project and am ready to start working. I guarantee quality and timely execution.

Show original

Translate

I am among the top 10 developers in the category of "Artificial Intelligence and Machine Learning" among ~2100 specialists on the platform.
I guarantee:
- Fast and quality execution of the task
- Strict adherence to deadlines
- Regular communication throughout the entire process
I would be happy to discuss the details of your project in private messages.

Show original

Translate

Hello. I have been working with React/Node.js for over 9+ years. I am ready for collaboration.

Show original

Translate