Ten years of AI that held in production.
Defence first, then finance, then six years inside one industry. Eight systems from that decade are rebuilt here in the open — designed to be read, and run on request.
It never claims
what it cannot show.
Every system here is labelled by how real it is — what ran in production, what is synthetic, what is simulated. If the label is missing, the claim is not made.
Architecture
The design, models and control flow that ran in production — rebuilt at a scale one machine can run.
Data
Synthetic, always. Generated sets that reproduce the real distributions; no customer data anywhere.
Integrations
Mail servers, cameras, ticketing — stand‑ins that are labelled as stand‑ins, in the demo and the docs.
Demo
Opened personally, for a defined period, to those evaluating the work. Everything else is public.
The eight, with the evidence.
Hover a row to see it run. Every system opens to its walkthrough, documentation and demo.
| # | System | Reality | Now |
|---|---|---|---|
| 01 | Enterprise Retrieval PlatformHospitality group · 2024 – 2025 | downscaled reimplementation · synthetic | case study · docs |
| 02 | Contract IntelligenceTrade finance · 2019 – 2020 | downscaled reimplementation · synthetic | case study · docs |
| 03 | Service IntelligenceHospitality group · 2024 – 2025 | real pipeline shape · synthetic | case study · docs |
| 04 | Multilingual CorrespondenceHospitality group · 2020 – present | real architecture · synthetic | docs |
| 05 | Personalisation & RecommendationHospitality group · 2020 – present | real architecture · synthetic | docs |
| 06 | Content IntegrityHospitality group · 2020 – present | real architecture · synthetic | docs |
| 07 | Generative EditorialHospitality group · 2020 – present | real architecture · synthetic | docs |
| 08 | Perception SystemsDefence electronics · 2016 – 2019 | downscaled reimplementation · synthetic | docs |
Where the work comes from.
I started where failure is expensive, learned that trust is an architecture concern, and then spent six years inside a single industry until I understood it end to end.
Abhishek Saxena architects and delivers AI platforms for enterprises — from defence‑grade perception to regulated document intelligence to the systems behind a European hospitality group.
Full profile →vLLM · Qdrant · Kafka · Redis
Postgres · TensorRT · ONNX Runtime
Kubernetes · Terraform · OpenTelemetry
TOGAF ADM · documents before code
Certified. Verifiable.
Smaller things, built for real use.
Reduced public edition of a product
One line: what the full product does for whom, and what this cut‑down public edition lets a visitor try.
live →An internal tool that saved a team time
One line: the job it did, the people it did it for, the measurable result if there is one.
write‑up →A proof of concept that answered a question
One line: the question, the answer, and what happened next.
code →A side project built to learn something
One line: what it is and what it taught.
live →Three steps. The same for every system — and for this site.
Scope
What, why, and what “done” means — written before anything is built. Documents before code, numbered gates, a decision log.
“We write it down before we build it.”
Build, in the open
Bounded chunks with a coding agent; one chunk, one branch. Everything labelled by how real it is.
“Demos you can click from the first week.”
Prove
A person reviews against a written quality bar, then a real browser — desktop, phone, reduced motion — before it ships.
“If it doesn’t pass, it doesn’t ship.”
The live systems are opened personally.
Documentation and walkthroughs are open to everyone. Access to the running environments is granted for a defined period, to those evaluating the work.
Request access →