§ 01 / 07The Thesis
measurementdepth⟨ data , model ⟩ = 00,0

The next frontier
of enterprise AI is
boring & measurable.

We are working on the frontier of AI-native data tooling, retrieval engines and optimized models that help enterprise AI move from impressive demos to reliable, auditable, ROI-positive production.

§ 02 / The Manifesto90°

We build AI to turn hidden knowledge
into verifiable intelligence.

AI is not a silver bullet. As models become easier to access, the real advantage moves to the systems around them: data, search, retrieval, verification, logging, cost, and deployment. 90degrees Labs builds that measurable layer between hidden enterprise knowledge and production AI.

01Why 90degrees?

Hype is easy. Value has to be engineered.

An agent calling tools is not a business result. A prototype is not a production system. AI only matters when it improves a workflow, reduces cost, increases quality, or helps people make better decisions repeatedly.

02Why data, search, and retrieval?

The model is only as useful as the data it can access.

As frontier LLMs become more and more arbitrary, advantage moves to the systems around them where commoditized intelligence is turned into competitive advantage.

03Why measurable?

Trust, but verify. Production AI needs evidence.

Teams must know what was retrieved, what was used, whether the answer was grounded, where sensitive data went, and how the system improved over time. What cannot be measured cannot be trusted at scale.

04Why boring?

Glamorous sells. Boring survives production.

Clean data, high-quality retrieval, small specialized models, consistent logging, low-latency deployment, optimized cost, and dedicated ownership are the unglamorous parts of AI that make it work — and make it pay.

§ 03 / 90degrees OS90°

Research the hard problems. Build the tools.
Test them in the wild. Deploy what works.

90degrees OS is how ideas move from open research to open-source, from technology to products, and from prototype to production systems with our customers.

  1. Research
    Think Deep

    Search 2.0 is an optimization problem in latent space: better retrieval, better context utilization, smaller specialized models, and indexes that improve from real-time feedback.

  2. Build
    Build Open

    We build and release infrastructure and models in the open so others can inspect, run, challenge, and build on the work.

  3. Test
    Eat Our Own Dogfood

    We turn the strongest ideas into products and test them with real users, real data, real failures, real latency, and real production constraints.

  4. Deploy
    Deploy Fast & Forward

    We work with enterprise teams to find the bottleneck, add measurement, fix the data layer, improve retrieval, reduce waste, and deploy systems that can actually be operated ROI-positive.

§ 04 / Open Research

Open Research

We research the systems around frontier LLMs: retrieval, small-model optimization, and runtime verification — the layer that makes AI grounded, auditable, and production-ready.

01/05
§012025.Q2 · paperPublished

LLM pruning and knowledge distillation on HPC.

Joint research endeavour with HLRS, AMD, and HPE on pruning and distilling LLMs at scale on AMD MI300A hardware inside Germany's secure national HPC.

  • HLRS
  • AMD
  • HPE
§02Enterprise searchOpen Source

ColSearch — document-native multi-vector retrieval.

High-performance multi-vector retrieval that treats the document, not the chunk, as the unit of search. Much simpler pipeline plumbing with a new quantization scheme for high efficiency-to-recall.

  • Multi-vector
  • Quantization
§03Research RepoOpen Source

LLM-Opt — pruning & knowledge distillation.

Structured pruning of LLMs and encoders, with distillation back into smaller, deployment-ready students. Built for cost, latency, and recoverable quality.

  • Pruning
  • Distillation
§04Open ModelsOpen Models

Open models on Hugging Face.

Specialist small models for the boring parts of production AI: compression, PII detection, relation extraction, agent routing, privacy-aware classification.

  • PaliGLiNERsoon
  • PrivacyPalisoon
§05Published · FFplus projectPublished

HALO — Hallucination-Aware Layered Optimization.

Dennis, our Co-Founder, set the foundation for LLM auditability at runtime during the EU-funded FFplus project.

  • Explainability
  • Sovereign AI
§ 05 / Products

Products

Production tools built on our open core. Each one targets a real bottleneck in enterprise AI — measured, named, and shipped.

01/04
§02RAG AgentsPublic Beta

TRACE

Real-time safety for knowledge agents.

TRACE protects knowledge agents from private-data leaks, unsupported answers, and wasted context — without replacing your stack.

Built onvllm-factory · latence-trace-python

§03DataPre-public · 2026

Data Intelligence Pipeline

Turn messy input data into clean markdown, structured extraction, and dataset-level knowledge graphs and ontologies.

Built onvllm-factory · ColSearch

§04MarketComing Soon

LatentFind — fixing the broken freelance market.

AI-native hire experience for freelancers and companies — without slow and expensive agencies in the middle.

Built onColSearch · TRACE

§ 06 / Consulting

Consulting

Forward-deployed AI engineering for enterprise teams. Each engagement ends with a decision you can act on — built on our open core.

01/05
000.0°Diagnose5–7 days

360° Analysis

A complete view of your AI surface — architecture, data, retrieval, controls.

DecisionKeep, replace, or retire — across stack and initiatives.

022.5°Audit5–10 days

AI Act Check

Translate AI Act requirements into operational technical controls — not paperwork.

DecisionWhere you comply, where you must close gaps, who owns what.

045.0°Evidence3–5 days

Data Proof Sprint

Prove what your data can actually support before committing to a bigger build.

DecisionBuild, replace, or do not build.

067.5°Rapid PoC2–4 weeks

FastForwardSprint

Design, prototype, and stress-test a scalable AI component on your real workload.

DecisionProductionize, harden, or throw away.

Built onvLLM Factory · TRACE · ColSearch

090.0°Build & OperateOngoing

Forward-Deployed Engineering

We embed in your team to build production workflows, agents, and retrieval — on our open core.

DecisionDay-to-day system shape, scope, and ownership.

Built onvLLM Factory · TRACE · ColSearch

§ 07 / Collaborate with us90°

A movement for boring, measurable AI.

Enterprise AI adoption does not need more hype. It deserves systems that can be evaluated, deployed, audited, optimized, and trusted.

90degrees Labs exists to accelerate that shift: from demos to production, from generic AI promises to company-specific systems, from black-box answers to measurable evidence.

90°
01 · For enterprises

Build AI systems that work on your data and survive production constraints.

90°
02 · For integration partners

Bring real-time compliance for RAG, multimodal high-performance retrieval engines, and small-model optimization to customers. White label and customizable.

90°
03 · For investors and collaborators

Join a focused lab building the search, retrieval, and optimization layer around enterprise AI.

Make your AI system measurable.

If your enterprise AI initiative depends on company data, search, retrieval, or RAG quality, 90degrees Labs can help you find the bottleneck and turn it into measurable progress.

Explore TRACE