Technology reference

The detailed view, for when you need it.

This is a working reference rather than a sales pitch. Use it to explore the technologies and architectural contexts behind our practice.

Architectural trade-offs

There is no universal best stack.

Compare a few common contexts and the considerations that shape an appropriate technical direction.

Architectural Trade-Off Analysis

We don't have a default stack.

The right technology depends on the problem, constraints, scale, and organizational lifecycle.

Select an Architectural Context

System Requirement WeightPRIORITY ALLOCATION

Speed to Market

Scale

Control

AI Capability

Cost Efficiency

Security

Existing Ecosystem

Long-Term Maintainability

Evaluation Scale: Tier 1 to 5Deterministic Scoring

Architectural ImplementationVALIDATED STACK

A high-velocity software product where shipping verified value quickly with strict typing and low operational maintenance is the overriding goal.

Frontend / ExperienceNext.js (App Router) + TypeScript + Tailwind CSS

Domain / Application LogicStrictly Typed Server Actions & RESTful Edge Routes

Transactional StorePostgreSQL (ACID-compliant with Prisma / Drizzle ORM)

Infrastructure & HostingManaged Cloud (Vercel / Cloudflare) + Managed Postgres

ARCHITECTURAL RATIONALE

Maximizes velocity by consolidating client and server boundaries in TypeScript, reducing deployment friction while maintaining a bulletproof relational foundation for future scale.

Zero default dogmatismContext over hype

Technology inventory

Areas we work in and evaluate.

The labels distinguish technologies we use regularly from areas where we have experience, can support a team, or are actively evaluating.

Technology Inventory & Evaluation

The Technology Explorer

A transparent architectural inventory. We categorize technologies by how we actually use and evaluate them in production—not by marketing claims.

GOVERNANCE TIERS:COREEXPERIENCEDAVAILABLEEVALUATED

Artificial Intelligence: Cognitive reasoning, neural models, agentic tool-use, and private inference infrastructure.

Filter Tier:

Foundation & Reasoning Models

7 COMPONENTS

OpenAI (GPT-4o / o3)

Core

Complex multi-step reasoning, structured JSON outputs, and fast general logic.

Evaluated for deterministic reasoning and high token reliability.

Anthropic (Claude 3.5 / 3.7)

Core

Long-context analysis (200k+ tokens), deep synthesis, and code generation.

Benchmark leader for agentic tooling and document evaluation.

Google Gemini (1.5 / 2.0)

Experienced

Extreme context windows (up to 2M tokens) and native multimodal audio/video understanding.

Selected for large archive auditing and video processing.

Meta Llama (3.1 / 3.3)

Core

Self-hosted sovereign open-source model execution within private VPCs.

Zero data leakage; zero per-token third-party dependency.

Mistral (Large / Codestral)

Experienced

Cost-efficient European sovereign hosting and fast latency inference.

Excellent performance-to-cost ratio for mid-complexity tasks.

DeepSeek (V3 / R1)

Evaluated

Advanced mathematical and structured logic benchmarking in private labs.

Under active lab verification for high-efficiency private inference.

Specialized Domain Models

Available

Bioinformatics, financial tabular reasoning, and legal clause extraction.

Introduced when generalized frontier models prove inefficient.

AI Engineering & Agentic Workflows

6 COMPONENTS

Agentic Systems & Tool Calling

Core

Multi-step autonomous execution with discrete software tool access.

Strict schema contracts with defensive error boundaries.

RAG & Semantic Retrieval

Core

Anchoring generative outputs to live enterprise relational & document stores.

Hybrid dense/sparse vector search with re-ranking.

Deterministic Structured Outputs

Core

Enforcing strict JSON Schema / Pydantic validation on model responses.

Eliminates hallucinated formats before downstream persistence.

Model Routing & Fallbacks

Core

Dynamic cost and latency routing with graceful multi-model failover.

Routes simple queries to fast models and complex queries to frontier models.

AI Observability & Tracing

Core

Per-token cost tracking, latency histograms, and semantic drift auditing.

Full lineage and reproducible synthetic benchmarks.

Human-In-The-Loop Systems

Core

Operator review queues for high-consequence automated decisions.

Confidence threshold gating prevents unmonitored risk.

AI Infrastructure & Serving

3 COMPONENTS

Vector Databases & Indexes

Core

pgvector, OpenSearch, and Qdrant for semantic similarity retrieval.

Integrated directly into PostgreSQL or distributed clusters.

Private GPU Model Serving

Experienced

vLLM and TensorRT-LLM on dedicated cloud or on-premise hardware.

Sub-second token throughput for private enterprise VPCs.

Knowledge Graph Integration

Available

Graph-augmented retrieval combining relational nodes with vector distance.

Used in complex organizational topology investigations.

Evaluated by architectural suitability, not marketing popularity.Zero vendor lock-in

Back to the technology overview