ITAD Assist/IT asset disposition
Find the best next life for every retired IT asset.
A decision-support tool that estimates reuse, refurbishment, component-harvest, material-recovery and disposal value for retired IT hardware — with a deterministic calculation engine an agent can consult but never overrule.
Recovery routes
One retired asset resolves into the route with the most remaining value.
- Sector
- IT asset disposition
- Year
- 2026
- Disciplines
- Agentic systemDecision supportDeterministic engineMCP tools
The problem
When an organisation retires IT hardware, the value is spread across several possible fates — resale, refurbishment, harvesting components, recovering raw materials, or safe disposal — and the right choice differs for every asset. That decision is usually made on intuition or a spreadsheet, with no auditable basis and no consistent way to compare routes or recycler quotes.
The approach
We separated judgement from arithmetic. A pure, decimal-only domain package holds the formulae; the calculations are deterministic and versioned, so every estimate freezes the assumption set and formula version it was produced under. Updating a commodity price later never rewrites a stored estimate.
Around that engine we built a client-facing application backed by a Python service: batches, CSV import, per-asset estimates, scenario overrides, quote comparison and recorded outcomes — all scoped to the organisation rather than to a request, so tenants stay isolated.
An agentic layer connects through MCP so an assistant can interpret messy asset descriptions, compare routes and explain an estimate in plain language — while the numbers stay owned by the deterministic engine, not the model.
What we built
- 01
Route estimation
Estimate value across reuse, refurbishment, component harvest, material recovery and disposal for each asset.
- 02
Deterministic engine
Decimal-only formulae in a pure domain package — auditable calculations with no floating-point drift.
- 03
Versioned assumptions
Every estimate freezes its assumption set and formula version, so a later price change cannot silently rewrite history.
- 04
Batches and import
CSV upload and organisation-scoped asset records, with frozen estimate snapshots per batch.
- 05
Scenarios and quotes
Compare route overrides and recycler quotes side by side without mutating the published assumption set.
- 06
Agent access via MCP
An assistant can identify assets, compare routes and explain an estimate — without owning the maths.
Technology
- SvelteKit BFF
- Python / FastAPI service
- MongoDB
- Decimal domain engine
- MCP (Streamable HTTP)
- Nx monorepo
Outcome
A working client product: an estimating workspace where an operator can import a batch of assets, produce auditable value estimates across every disposition route, compare quotes and record outcomes — with an agent able to assist through the same service.
Seeded prices, yields and material compositions in the build are illustrative, not live market data.