AI architecture for automatic generation of astrological reports

Budget: 8000 UAH Deadline: 5 days

Hello, Kateryna. The key in your project is that the report should not be a "AI-style" canvas of 25 pages. This is solved by architecture, which is what I am working on.

Here’s how I see the architecture:
• Formulas and calculations — a separate deterministic code module, not LLM. The model does not calculate — this way the result is accurate and reproducible.
• We digitize the knowledge base (your books, interpretations) into a structured format, where each interpretation is linked to the result of the formula — not a "dump of text."
• The report is assembled section by section: the master prompt sets the structure and voice, individual section prompts write prose from already prepared data. This results in 20–25 pages of stable premium quality without hallucinations and without context overflow.
• "Non-template nature" — this is the canon of structure and tone that the model adheres to strictly. This approach I built for the client content brand EverStory: the canon as a single source of truth eliminates the generic-AI appearance.
• PDF: HTML/CSS template, headless render in a premium document.

Similar AI projects:
• EverStory — an automated AI pipeline for content generation that I built for the client (everstory.pro).
• ODIS — an AI agent on Claude that conducts dialogue and delivers structured results (odis-group.vercel.app).

Stack: Python (formula module + orchestration), Claude (text composition), PostgreSQL + pgvector for the knowledge base, HTML/CSS, headless render for PDF.

The main factor affecting the timeline is the current state of your knowledge base. One question: are your formulas and interpretations currently structured (tables, documents) or are they still "raw" books and texts? This will determine the accurate assessment. Expand

Budget: 6000 UAH Deadline: 5 days

Hello! This is a very interesting project that would be nice to work on. The main thing here is to properly organize your methodology into a system, rather than just "plugging in AI." The formulas need to be calculated separately and accurately, the knowledge base must be structured, and based on this, AI will write a personal report.
Regarding the architecture: the formulas are calculated by a separate deterministic Python module without the involvement of LLM, the knowledge base (your books, interpretations) is digitized in such a way that each interpretation is tied to a specific calculation result. The report is assembled section by section — the master prompt sets the structure and tone, and individual section prompts write prose from already prepared data. This eliminates template-like responses and hallucinations.
Stack: FastAPI on the backend, Claude API for generation, ChromaDB for vector search in the interpretation database, WeasyPrint for PDF templates, SendGrid for email delivery. Timeline — approximately 3–4 weeks.

Budget: 7000 UAH Deadline: 10 days

Hello. The task is interesting, and I would gladly look deeper into it.

In my opinion, the main thing here is not just to connect AI, but to properly organize your methodology into a system. That is, formulas need to be calculated separately and accurately, the knowledge base should be well-structured, and the AI should then write a good personal report based on this, without making unnecessary assumptions.

I wouldn't do everything in one big prompt. It's more reliable to break the process into several steps: client data → calculations → selection of interpretations → generation of report sections → structure verification → PDF → email.

From a technical standpoint, this could be a backend in Python or Node.js, a knowledge base + vector search, OpenAI/Claude/OpenRouter for text generation, a separate module for formulas, and HTML/PDF templates for a nice final report.

Regarding timelines: it can be done in 1-2 weeks.

For an accurate estimate, I need to see a sample report, your calculation structure, and part of the materials from which the knowledge base should be formed. Expand

Budget: 5000 UAH Deadline: 30 days

Hello!

You are describing a system that I have already implemented — astro-synth.com:

— the user inputs data
— the system performs calculations
— AI generates a personalized PDF report
— automatic delivery to email

This is a live project.

Here’s how I see the architecture of your system:

1. Knowledge Base (RAG): we digitize your books, formulas, and interpretations → vector database → AI refers only to your methodology, does not invent

2. Logic of Formulas: each formula is a separate module, the result is passed to the prompt as structured context

3. Master prompts: a system of mega-prompts where each section of the report is generated separately → stable result without "AI-style"

4. PDF Generation: premium layout, 20-25 pages, branded style

5. Scalability: architecture ready for connecting different LLMs (GPT-4o, Claude, local models)

Stack: PHP/WordPress, OpenAI/Claude API, RAG (Pinecone or Supabase pgvector), PDF generation.

Please send a sample report and structure — I will provide a specific estimate. Expand

Budget: 5000 UAH Deadline: 5 days

Hello,

I understood the specifications: personalized 20–25 page PDF reports generated through LLM based on your knowledge base (formulas, methodologies, interpretations), and delivered to the client via email after data entry.

Here’s how I see the architecture (implementation order):
1. **Knowledge base → structure.** Your texts, books, interpretations — digitized into a structured form: YAML/JSON blocks with tags (planet, aspect, number, context). This is the foundation — without it, even the best LLM prompt will be "AI-style."
2. **Formulas and calculations** — a separate Python module, fully deterministic, without LLM. Input — client data, output — structured result (natal chart, numbers, etc.). There should be no room for hallucinations here.
3. **RAG layer** — pgvector (Postgres) on top of your knowledge base, filtering by tags before vector search, so LLM only takes relevant interpretations.
4. **Master prompt + section prompts** — Claude Opus with prompt caching (saves up to ×10 on cost), separate prompts for report sections, JSON output, validation on our side before PDF assembly.
5. **PDF** — WeasyPrint with HTML+CSS template of your corporate style (more reliable than ReportLab for long documents with graphics).
6. **Delivery** — Resend or SES, generation + deletion of the file after 24 hours. Expand

Budget: 5000 UAH Deadline: 1 day

Good day. I am ready to complete this project as I have extensive experience in app development.

Budget: 5000 UAH Deadline: 10 days

Good day. I have experience in building AI systems for generating personalized content, working with LLM, prompt architecture, RAG, and automating complex report generation logic. I can show similar projects during our conversation.

I envision the implementation as follows: first, a structured knowledge base is formed from your materials, after which the logic of calculations and relationships between parameters is built. Next, a multi-level prompt system with separate roles and mega-prompts for different parts of the report is created. For stability and scalability, it is advisable to use RAG + vector database, so that the AI operates based on your methodology rather than "inventing" interpretations.

The system can automatically:
- accept user data,
- perform calculations,
- generate a structured premium report,
- create a PDF and send it to email without manual intervention.

For the tech stack, I usually use:
- Python / Node.js,
- OpenAI / Claude API,
- LangChain / RAG,
- Pinecone or Weaviate,
- PostgreSQL,
- PDF generation pipelines,
- workflow automation.

The timelines depend on the scope of the methodology and the amount of logic, but such a project is better built step by step — from the MVP core of generation to a full production system.

I would be interested to see an example of your materials and the structure of the reports to immediately assess the optimal architecture and scope of work. Expand

Budget: 5000 UAH Deadline: 1 day

Hello! Your project looks very strong and promising—especially because you already have a ready methodology, formulas, and a base of interpretations. This is exactly the foundation on which a truly premium AI system can be built, rather than just another template "text generator."

I specialize in AI systems architecture, Prompt Engineering, and building RAG/LLM solutions for personalized content generation. What you describe, I see as a full-fledged intelligent platform with multiple levels of logic:

— structured knowledge base
— formula and calculation engine
— layer of interpretations
— AI orchestration / master prompts
— generation of quality narrative reports
— automatic assembly of premium PDFs
— email delivery pipeline

How I see the architecture:

1. Knowledge Base Layer
Digitization and structuring of your materials:
books, texts, interpretations, videos, rules, relationships between parameters.

2. Formula & Logic Engine
A separate module with verifiable formulas and logic of calculations, so that AI does not "invent" values but works only based on valid data.

3. RAG + Semantic Retrieval
AI receives only relevant interpretations and context for a specific client.
This helps avoid "fluff" and template-like responses.

4. Master Prompt System
Building multi-level mega prompts:
— tone of voice
— report structure
— depth of analysis
— logic of sections
— anti-AI-style control
— consistency checks

5. Report Generation Pipeline
Generating the report in blocks → checking coherence → final assembly → PDF generation.

6. Scalability
The system is designed from the outset to be scalable in the future:
— OpenAI / Claude / Gemini
— multilingual reports
— different types of reports
— API / web platform
— automation and high-load generation

For the stack, I usually use:
— OpenAI / Claude APIs
— Python / Node.js
— LangChain / LlamaIndex
— Vector DB (Pinecone / Weaviate / Qdrant)
— PostgreSQL
— PDF generation pipelines
— automation orchestration

From similar experience:
— AI systems for personalized reports
— generation of expert long-form content
— RAG platforms
— AI knowledge systems
— prompt orchestration for structured documents

Regarding timelines:
An MVP of such a system usually takes about 4–8 weeks depending on the volume of the knowledge base and the complexity of the formulas.
After the MVP, the quality of interpretations and automation can be gradually enhanced.

I particularly like that you want to create a quality premium product. Expand

Budget: 5000 UAH Deadline: 3 days

Hello! I have reviewed the project and am ready to start working. I am confident you will be satisfied with the result.

Budget: 5000 UAH Deadline: 14 days

The task is clear: take your methodology, digitize the knowledge into a vector database (RAG), build a prompt system with mega-prompts that pulls relevant interpretations for a specific profile, and output a structured PDF of 20+ pages without an "AI tone".

Stack: OpenAI/Claude + LangChain + Pinecone/Chroma, PDF via WeasyPrint or Puppeteer, automation through n8n or FastAPI pipeline.

I will immediately build the architecture to be scalable for connecting new LLM models.

How many unique types of reports are planned at the start — one (for example, a natal chart) or several in parallel?