Capabilities
Nothing here is aspirational. Each capability links to the delivered project where it was proven.
On-prem & air-gapped AI
- Self-hosted LLM serving on commodity CPU-only hardware (llama.cpp, quantised GGUF)
- Grounded, citation-backed summarisation with hallucination-refusal guardrails
- Offline, image-based installers with backup and one-command rollback
- Air-gapped release engineering: USB delivery, sha256 packs, idempotent installs
Evidence
Regulated platforms & GovTech
- Treasury payment-gateway integration with server-side reconciliation
- Configuration-driven policy engines — statutory values as data, not code
- Atomic, gap-free statutory numbering proven under concurrency
- Multi-role approval workflows, KYB verification, audit-by-design schemas
- WCAG 2.1 AA / GIGW accessibility for government portals
Evidence
AI intelligence & vision systems
- Multi-source evidence triangulation (documents + reports + call recordings)
- Explainable visual search: 733-signal matching over 70,000-item catalogues
- Multi-provider LLM abstraction — swap models without a code deployment
- Local embeddings and vector search with zero data egress
Evidence
Delivery engineering
- Five-layer embedded testing: unit, deep-smoke, synthetic monitoring, log/LLM triage, confidence gating
- Two-gate completion: automated checks and verified browser behaviour, both required
- Four-environment release pipelines with production safety guardrails
- AI-augmented lean delivery: founder + agent fleet in place of a conventional team
Evidence