LOCKEDIN / FORWARD-DEPLOYED ENGINEERING 101
Understand it.
Then build it.
You can be an experienced engineer and still be new to agents. Start with what each thing is and how it helps a person do useful work. Explore 52 original introductions across software, agents, cloud, security and delivery. Start here without a model key; each topic connects to deeper instruction and practice.
- 101UnderstandPlain language, visual walkthrough, vocabulary
- 102Build with guidanceOriginal lessons, examples and knowledge checks
- 103Practise and explainRunnable labs, failure cases and handover
Data foundations · object storage, SQL and retrieval
Files, databases and search indexes: what goes where?
An application needs somewhere to keep information and a way to find it again. Object storage keeps named objects such as documents and images. A relational database organizes records into tables you can query and update. A search index helps find relevant material. These are different jobs, even when one product can do more than one of them.
An FDE needs to explain where the customer's information lives before connecting it to an agent. A convincing answer is not enough if it used an outdated procedure, lost a ticket update or crossed a customer's access boundary.
UNDERSTAND THE SYSTEM
One assistant, three different data jobs
These sample records describe one fictional system. The parts view shows the starting snapshot. Select a part to see what it stores. Follow the example to see why an old search result cannot become current evidence just because it looks relevant.
Read the connections in plain English
- 01Procedure document → Searchable copy
Ingestion creates a derived search representation. An update or deletion must be propagated deliberately; the copy can lag behind the source.
- 02Assistant app → Ticket record
The app issues a scoped query for ticket T-104. The database owns that ticket's current status.
- 03Assistant app → Searchable copy
The app searches only the caller's permitted source scope. Results are candidates, not established facts.
- 04Searchable copy → Evidence check
The candidate carries its source ID and revision into an application check. A plausible match is not enough.
- 05Procedure document → Evidence check
This illustration's trusted source metadata establishes the current revision and access. Existence, freshness and permission all matter before model context is built.
- 06Evidence check → Model context
Admit only a current, permitted candidate. Exclude stale, denied or deleted-source entries; report missing evidence when nothing usable remains.
Check your understanding: If the procedure is deleted, can the remaining index entry still be used?
No. A leftover search copy is not evidence that its source still exists or that access remains valid. Exclude it from new model context, remove derived entries through the deletion process, and handle caches and backups under the explicit retention policy. Deleting one copy is not proof that all copies are gone.
CONNECT IT TO WHAT YOU KNOW
Start with something familiar.
Think of a shared drive containing manuals, a spreadsheet tracking tickets and a catalog that helps you find a manual. Each holds information for a different reason. The application connects them using stable identifiers, just as a ticket can reference a particular manual.
A database adds queries, constraints and transaction behavior beyond the spreadsheet analogy. An index can live inside a database or in a separate service. Searchable copies are not automatically current, and finding a document does not grant permission to read it.
Words worth knowing
- Object / key
- Stored bytes and metadata, retrieved by a name. The data can contain a structured format such as JSON; object storage does not itself make it a relational table.
- Table / row / primary key
- A table groups records with named columns. A row is one record. Its primary key is the unique identifier, such as T-104.
- SQL / transaction
- SQL is a language for working with relational data. A transaction groups changes so they can commit or roll back together.
- Search index / embedding
- An index supports finding data. An embedding is a numerical representation used by vector search; it is not the original document, an access decision or proof of truth.
- Source of truth / derived copy
- The authoritative place for a particular fact versus information built from it. Decide this per fact: ticket state can belong in the database while procedure content belongs in an object.
- Block storage / filesystem
- A disk exposes blocks that an operating system can organize into files. It can hold database files, but the database software supplies queries and transactions. A disk is not an object-storage bucket.
ONE EXAMPLE / FICTIONAL CASE
Follow the connections.
Field Desk is a fictional maintenance assistant. It keeps procedure P-17 as a document, ticket T-104 as a structured record, and a searchable copy of the procedure. A technician updates P-17 from revision 1 to revision 2. The search copy has not caught up yet.
UNDERSTAND THE SYSTEM
One assistant, three different data jobs
These sample records describe one fictional system. The parts view shows the starting snapshot. Select a part to see what it stores. Follow the example to see why an old search result cannot become current evidence just because it looks relevant.
See what moves and why.
Read the connections in plain English
- 01Procedure document → Searchable copy
Ingestion creates a derived search representation. An update or deletion must be propagated deliberately; the copy can lag behind the source.
- 02Assistant app → Ticket record
The app issues a scoped query for ticket T-104. The database owns that ticket's current status.
- 03Assistant app → Searchable copy
The app searches only the caller's permitted source scope. Results are candidates, not established facts.
- 04Searchable copy → Evidence check
The candidate carries its source ID and revision into an application check. A plausible match is not enough.
- 05Procedure document → Evidence check
This illustration's trusted source metadata establishes the current revision and access. Existence, freshness and permission all matter before model context is built.
- 06Evidence check → Model context
Admit only a current, permitted candidate. Exclude stale, denied or deleted-source entries; report missing evidence when nothing usable remains.
Check your understanding: If the procedure is deleted, can the remaining index entry still be used?
No. A leftover search copy is not evidence that its source still exists or that access remains valid. Exclude it from new model context, remove derived entries through the deletion process, and handle caches and backups under the explicit retention policy. Deleting one copy is not proof that all copies are gone.
FROM RECOGNITION TO REASONING
Place three pieces of information, then catch a stale copy.
- 01
Assign a procedure PDF, a ticket's changing status and searchable procedure excerpts to their respective roles. Explain what question each store answers.
- 02
Give the procedure ID P-17 and the ticket ID T-104. Write one row with columns ticket_id, status and procedure_id. Name which source owns each fact.
- 03
The document now says revision 2, but search still returns revision 1. Predict whether that candidate may enter the model's context, then follow the visual example.
- 04
Open the free lab in Go deeper. Run its small SQLite query, refresh the example index, then revoke access and delete the source. Explain why a leftover search entry must not reach the model.
A three-store data map, a ticket query and a four-case table: stale, current, denied and deleted.
Compare your reasoning when you are ready
The object holds the procedure, the table owns ticket status, and the index helps locate candidates. Matching a source ID is insufficient: current revision and access must also agree. Refreshing the index fixes staleness but cannot fix denied access. Deleting the source requires handling derived copies and retention separately.
Keep notes in your own working copy. Moving through these cards does not record completion.
TURN THE IDEA INTO USEFUL WORK
You can now go deeper.
Start with the question: retrieve a file, update a fact or find relevant material? Then name the authoritative source, access rules and update path. Learn retrieval and embeddings after these storage responsibilities are clear.
Academy lessons use your existing training access. Free guides stay open.