No client logos. No invented numbers. Judge the engineering.
Most agency portfolios are a list of names and a percentage. Neither tells you whether the team can build the thing you need. So we publish the shape of the work instead.
JANNEX does not publish client names, logos, testimonials or outcome metrics. The projects below are our own demonstration builds and reference methods, described so you can judge the engineering rather than the marketing. Where a real engagement is relevant to your problem, we will discuss it directly with you under the confidentiality it deserves.
Knowledge assistant over private documents
A retrieval assistant that answers from an organisation's own policy and process documents, cites the paragraph it used, and reports every question it could not answer.
Staff questions are answered by a handful of people who know where the documents are. The documents themselves are current in some places and three revisions old in others.
Permission-aware ingestion, hybrid keyword and vector retrieval, answers constrained to retrieved text with citations, and a confidence threshold below which the assistant declines and offers a route to a person. Unanswered questions accumulate into a content backlog.
- Ingestion respects source permissions
- Every answer carries a citation
- Declines below a confidence threshold
- Gap report drives content updates
- Evaluation set run before each release
Bounded agent for back-office exceptions
An agent that resolves routine exceptions across two systems, verifies its own output against business rules, and escalates anything outside its permitted boundary.
A recurring reconciliation task consumes hours daily. Most cases are mechanical; a minority genuinely need judgement, and the two are mixed together.
An explicit tool catalogue with scoped credentials, idempotent writes, a verification stage against schema and business rules, an approval gate on anything financial, and a readable trace of every run including cost.
- Read and write permissions separated per tool
- Writes staged or reversible
- Verification before commit
- Human approval on financial actions
- Per-run cost and action trace
AWS cost and architecture review
The review method we run against an AWS estate: architecture read alongside the bill, waste attributed to an owner, and changes sequenced from zero-risk to structural.
Spend is rising faster than usage and cannot be attributed to a team, product or customer. Nobody is confident which resources are safe to remove.
Cost and Usage Report analysis against measured utilisation, an inventory of orphaned and idle resources mapped to owners, storage lifecycle and retention review, pricing-model modelling, then a tagging standard, budgets and anomaly alerts so the result holds.
- Read-only access to produce findings
- Each change costed from your own usage data
- Sequenced by risk, not by size
- Rollback path per change
- Monthly review rhythm established
What we can show you privately
In a conversation, and where the client permits it, we can walk through architecture decisions, code, delivery history and the parts that went wrong.
That is a more useful evaluation than a logo wall, and it is the one we would want if we were choosing an engineering partner. It also means what we build for you stays yours to disclose.
Ask for a technical walkthroughHave something worth building?
Tell us the constraint you are working against. If we are not the right people for it, we will say so.
Or write to connect@jannex.in