EN / RO

VoidXD Full-stack platforms and cost-aware LLM systems.

We build multi-user platforms, internal tools, automation pipelines, and LLM workflows — and we keep them running. Practical engineering, direct communication, no agency overhead.

Based in Bucharest, working across EU timezones.

What we build

Two practices, one team. Both shipped to production and maintained.

01

Full-stack platforms

Multi-user web applications, dashboards, admin panels, portals, internal tools, roles and permissions, APIs, reporting systems, and database-backed workflows.

02

Data, automation, LLM pipelines

Scraping, data extraction, enrichment, deduplication, classification, document processing, monitoring, API workflows, and LLM-assisted operations.

An enrichment pipeline, instrumented

One trace from a typical record-enrichment agent — fetch, classify, extract, validate, summarize. Every step is observable, attributable, and budgeted.

Success trace_9fd7c1b2 run: company_enrichment
VOIDXD SRL Enrichment
Input: "Enrich VOIDXD SRL from website + public web results"
Latency 3.45s
Tokens 4.8k
Cost $0.003
Quality 0.94
Run enrichment pipeline root 3.5s
lead.enrichment.run
Orchestrates scraping, parsing, and LLM extraction for a single company record.
Find official sources tool 182ms
discover.urls
Search the web for official domains and social profiles.
Scrape selected pages tool 648ms
fetch.batch
Fetch 3 target URLs in parallel.
Fetch company homepage 211ms · 48KB
fetch.website.home
Retrieved homepage (200 OK).
LinkedIn (rate limited) warn 88ms
fetch.linkedin
Rate limited — falling back to cached data.
Clean and extract text parse 524ms
extract.content
Strip boilerplate, convert HTML to markdown for the LLM context window.
Classify and summarize llm 894ms
llm.classify_and_summarize
gemini-flash-latest
Extract business details, summarize value proposition, classify industry vertical.
Score lead confidence llmqa 312ms
llm.score_lead
gemini-flash-lite-latest
Assign quality tier and confidence score against ideal customer profiles.
Return structured result 77ms
response.serialize
Format JSON output, attach trace metadata, return 200 OK.

Problems we usually solve

Six shapes a project usually takes.

  • 01 You need a business platform, portal, dashboard, or internal tool.
  • 02 You have a manual process that should become software.
  • 03 You have data scattered across websites, spreadsheets, APIs, or documents.
  • 04 You need extraction, enrichment, classification, or reporting pipelines.
  • 05 You want to use LLMs, but cost, reliability, and accuracy matter.
  • 06 You have an existing system that is slow, fragile, or hard to maintain.

How we work

Four steps, every project.

  1. 01
    Understand the real problem

    We clarify what needs to happen, what can be ignored, and what would make the system useful in practice.

  2. 02
    Design the smallest useful system

    We prefer simple, maintainable architecture over overbuilt platforms.

  3. 03
    Build, test, and iterate

    We ship working software, then improve it based on real usage and feedback.

  4. 04
    Maintain or hand off

    We can keep improving the system or document it for your team.

Cost-aware LLM systems

We do not treat LLMs as magic. We design workflows around cost, latency, accuracy, and failure modes.

That can mean cheaper models for simple steps, stronger models only where needed, structured outputs, caching, batching, fallback logic, human review, and normal code instead of AI when normal code is enough.

  • structured outputs
  • caching
  • batching
  • fallbacks
  • human review
  • model routing

Common work

Where we are most useful.

Internal platforms

Dashboards, admin panels, multi-user tools, workflows, permissions.

Data enrichment

Scraping, APIs, spreadsheets, company data, classification, validation.

LLM pipelines

Extraction, summarization, routing, caching, cost control, observability.

Stabilization

Existing tools, fragile scripts, slow automations, messy integrations.

We are most useful when there is a real process behind the software, not just a page that needs to look nice.

FAQ

Short, honest answers.

Do you take fixed-price projects or hourly?
Both, depending on shape. New platforms usually go fixed-price per milestone after a short scoping pass. Maintenance, automation pipelines, and ongoing LLM work are typically retainer or hourly.
Do you work with non-technical founders?
Yes. We translate vague business problems into the smallest useful system, document assumptions, and push back on scope when it would make the project worse.
How do you handle LLM costs?
We design pipelines around cost from day one — model routing, structured outputs, caching, batching, fallbacks, and plain code where AI isn’t needed. We can also instrument cost-per-task tracking.
Can you take over an existing system?
Yes. Short audit, document what’s there, stabilize the painful parts, then iterate. We don’t rewrite for the sake of rewriting.

Start with the messy part

Send us the workflow, spreadsheet, scraper, dashboard, or LLM process you want turned into a reliable system. We'll tell you what should become software and what the first version looks like.

[email protected]