4Square Innovations® · AI from Vilnius

Cutting‑edge AI,
made actionable.

4Square Innovations UAB reads everything your business already runs on — ERP, CRM, documents, decisions — and turns it into intelligent automation. Cited. Governed. Ready to act on.

5 products · 24/7 automation · <2% hallucination · Vilnius · EU

Five products. One intelligent core.

Revolutionary AI solutions designed to transform how businesses operate and scale.

02 — Orchestration

Genius²

Multi‑LLM orchestration engine

Multi‑LLM orchestration with a consensus methodology that increases accuracy and reduces hallucinations in AI‑generated responses and decisions.

  • Advanced machine learning
  • Predictive analytics
  • Automated decision making
  • Real‑time data processing
  • Enterprise integration
Contact sales →

03 — Prompting

Prometheus

No‑code prompt management system

A no‑code prompt management system that continuously improves the results you get from AskDiana.ai through optimization and testing workflows.

  • No‑code prompt engineering
  • A/B testing for prompts
  • Performance analytics
  • Integration with AskDiana.ai
  • Automated optimization
Learn more →

04 — Compliance

Redact

REST API redaction & compliance tool

A REST API redaction tool supporting GDPR, HIPAA and CPA compliance — adjustable to identify and redact any information you consider confidential.

  • GDPR · HIPAA · CPA compliance
  • Custom redaction rules
  • REST API integration
  • Real‑time processing
  • Audit trail & logging
Learn more →

05 — Knowledge

Knowledge Decay

Intelligent knowledge lifecycle engine

Identifies the latest information in a data lake regardless of creation date — building knowledge pathways and depreciating older information automatically.

  • Intelligent content analysis
  • Knowledge pathway mapping
  • Automatic depreciation
  • Timeline‑independent discovery
  • Data lake optimization
Learn more →

Leading AI from the heart of Europe.

4Square Innovations UAB is a technology‑driven company based in Vilnius, Lithuania, and a member of the 4Square Capital Group. We build sophisticated AI that bridges the gap between complex technology and practical business applications.

Our team combines deep technical expertise with innovative thinking to create products that meet today's challenges and anticipate tomorrow's opportunities. We believe in the transformative power of AI to enhance human capabilities and drive sustainable growth.

With our location in Lithuania's thriving tech ecosystem and backing from 4Square Capital, we serve both European and Middle Eastern markets with world‑class AI.

5
Core products
24/7
AI support
Scalability
2024
Founded

Help shape the future of AI.

Access our comprehensive investor data room for detailed financial information, business plans, and growth projections.

Access data room →

Thinking harder about how LLMs think.

We investigate the structural failure modes that prevent LLMs from reasoning reliably over long contexts — and build theoretical frameworks and working implementations to address them.

01

Context Decay

The progressive reduction in a model's effective use of early context tokens as a conversation grows — silently, within the window. All tokens are present, yet earlier information exerts diminishing influence.

attention-asymmetry recency-bias positional-encoding

02

Epistemic Context Management

Context management is not a token budget problem — it is an epistemic structure problem. A conversation constructs a shared understanding with topology, temporal validity, density, and frame, grounded in nine cognitive-science memory models.

scoring-proxy cognitive-science tiered-retention

03

Dissolution Through Consensus

Instead of compressing the matter of a conversation, synthesise its boundary state. Our 'Forever Solution' processes history through a multi-model ensemble — the Genius² Digital Senate — replacing linear history with a present-tense Integrated Understanding.

genius² multi-llm boundary-synthesis

04

Bitemporal Context

Every statement carries two temporal attributes: when it was made (transaction time) and the interval during which it is true (valid time). Forward commitments should not decay by positional age.

valid-time transaction-time temporal-logic

05

Knowledge Taxonomy

Context decay interacts differently with vertical (deep, specialist) and horizontal (broad) knowledge. Vertical knowledge has no parametric fallback — when it degrades in context, it is simply gone.

vertical-knowledge horizontal-knowledge retention-policy

06

LLM Memory Architecture

Grounded in the Atkinson–Shiffrin multi-store model, Baddeley's episodic buffer, the serial position effect, and McGeoch's interference theory — mapping how human memory solves the same challenge LLMs face.

working-memory episodic-buffer interference-theory
Featured paper · 2026

Conversation Intelligence Architecture

A Theoretical Framework for Epistemic Context Management in Large Language Models — T. Larcombe

"Large language models manage conversation with policies derived from a single assumption: that context management is a token budget problem. This dissertation argues that assumption is wrong. Context management is an epistemic structure problem."
— Abstract
"What if the goal is not to preserve the conversation's content but the understanding it established? What persists is the shape of what was established. The boundary, not the matter."
— Chapter introduction

This work develops four original contributions: a cognitive‑science‑grounded scoring proxy, Dissolution Through Consensus, the four‑level Conversation Intelligence Architecture, and a bitemporal extension to context scoring.

Available under an open research licence — get in touch.

Four levels of the CIA

The Conversation Intelligence Architecture addresses four orthogonal dimensions of epistemic context management, each resolving a failure mode the previous level creates.

01

What to retain

Surprise-Proportional Allocation

Budget allocation proportional to prediction error — grounded in the Free Energy Principle. What the model did not already know is precisely what is worth retaining.

02

How to represent it

Navigational Topology

The established conceptual space should be a navigational map, not a content archive. A map of where knowledge lives is more useful than a catalogue of what it contains.

03

How to measure success

Generative Competence

Grounded in Kolmogorov complexity theory — the only valid test of a sufficient representation is whether it generates sufficiently. Surface similarity is not the right metric.

04

What threatens coherence

Adversarial Frame Agent

A parallel agent challenges the frame of the conversation, not its conclusions — surfacing the silent accumulation of unexamined premises that constitute the frame itself.

What the research shows

Context decay is silent

Unlike hard context window limits — which produce visible errors — context decay fails silently. The model continues to respond coherently while progressively disregarding early instructions and established facts.

>30% retrieval degradation

Facts placed in the middle of a long context degrade in retrieval accuracy by more than 30% relative to facts at either end — a U-shaped attention distribution that persists across model families (Liu et al., 2024).

Thinking tokens double the problem

Extended reasoning traces accumulate thinking tokens at roughly the same rate as conversational tokens, effectively doubling the rate at which earlier content is displaced — without the user being aware.

Dissolution achieves infinite horizon

By replacing linear history with a boundary state synthesised through multi-model consensus, Dissolution Through Consensus enables theoretically infinite conversational coherence.

Build real AI. Ship real products.

A hands‑on internship at the frontier of enterprise AI — working directly on production systems used by real clients, mentored personally by a CTO with 30 years of transformational technology leadership.

<2% Hallucination rate — Genius²
13+ LLMs queried in parallel
5 Live AI products
30yr CTO experience

Unpaid internship. These positions are unpaid. Successful interns receive a certificate of completion and a detailed personal reference from the CTO. We believe in being transparent about this from the start.

OPEN

AI / ML Engineering

Work directly on Genius² and AskDiana's RAG pipeline. Improve the consensus algorithm, evaluate embedding models, benchmark retrieval strategies, and experiment with LLM selection logic.

LLM Orchestration RAG Embeddings Vector DB Python asyncio
OPEN

Full-Stack Engineering

Build features across the AskDiana stack: Flask API endpoints, Next.js UI components, and the multi-tenant middleware layer. Real PRs against a real codebase with code review, migrations, and deployment pipelines.

Python/Flask Next.js 14 TypeScript MySQL Docker
OPEN

Data Science & Analytics

Design and run evaluations of RAG pipeline accuracy, benchmark consensus quality in Genius², and build predictive analytics features inside AskDiana's KPI module. You'll own the metrics that tell us whether our AI is actually working.

sklearn pandas Cosine Similarity TF-IDF Eval Frameworks
1 PLACE

DevOps & Infrastructure

Manage the infrastructure that keeps production AI running. Docker Compose deployments, Apache reverse proxies, Let's Encrypt, GPU-backed Ollama endpoints on RunPod, and observability via Langfuse.

Docker Apache RunPod GPU Langfuse Linux
1 PLACE

Product & Strategy

Work directly with the CTO on product direction, client integration playbooks, and market positioning. Research competitive intelligence, draft technical documentation, and support enterprise sales conversations.

Product Thinking Technical Writing Market Research Enterprise AI

Week 1–2

Onboarding & orientation

Deep-dive into the product stack. Set up your environment, run the products locally, read the architecture docs, and pair with a senior team member on your first issue.

Week 3–10

Delivery sprint

Contribute to the product backlog. Weekly 1:1 mentoring with Tony. Participate in architecture reviews and standups. You ship real features — not demos.

Week 11–12

Capstone & presentation

Present your body of work to the team. Document your contributions for the roadmap. Discuss paths to a longer-term role if there is mutual interest.

12 Weeks Duration
Remote or Vilnius Location
Unpaid Certificate & reference provided
Rolling Intake Apply early
Apply now → View programme →

Ready to transform your business?

Get in touch with our team to put governed, cited AI to work across your operations.

Headquarters

Veiverių g. 9B‑1
LT‑11346, Vilnius
Lithuania
European Union

Parent company

4Square Capital
SPC Free Zone
E311 Sheikh Mohammed Bin Zayed Rd
Sharjah, United Arab Emirates