Online professional certificate
RAG, Enterprise Search & Knowledge Systems
Build a small, testable search and grounded-answer feature from permitted sources. Compare retrieval methods, trace answer claims to exact passages, test access failures and hand measured results to decision owners.
- Format
- Online, self-paced
- Study time
- Up to 1 month
- Curriculum
- 20 applied lessons
- Language
- English
Practical capability
Build a search feature whose evidence and access can be checked.
Use one supplied enterprise case to turn an information need into permitted sources, judged retrieval, supported answers and a reviewable handoff.
Translate a vague request into a user question, an acceptance example and a safe stop.
Record authority, processing rights, source versions and identity-specific permissions.
Inspect extraction, design traceable passages and validate fields, filters and updates.
Compare lexical, semantic and hybrid results against fixed questions and access rules.
Build a small answer path that checks identity, exact source spans and citations.
Test support, revocation, freshness, delay and cost; route decisions to accountable owners.
Who this course is for
For professionals building reliable knowledge access.
A practical route for people who help prepare sources, implement retrieval, evaluate answers or support an enterprise search service.
The operating cycle
Move from a question to a tested answer path.
Follow the working sequence from the course case: define the task, verify source and identity boundaries, build retrieval, inspect evidence and hand off measured results.
Curriculum
Four modules. Twenty applied lessons.
Define the information task and its boundaries
Frame a support information need, identify authoritative sources, distinguish processing rights from user access, and route unresolved decisions to the right owner.
01Clarify the search task
Define an information task and measurable success and stop conditions with a domain owner.
Five practical steps
- Name the support user and decision
- Collect one exact sample question
- Identify expected source and known wrong answer
- Set an observable success and stop case
- Route open terms to the domain owner
Primary deliverable: Information-task brief.
02Establish source authority and permitted use
Record authoritative sources and the processing modes permitted for each.
Five practical steps
- Inventory the six case sources
- Name each source owner and current version
- Check permitted processing modes
- Hold unknown copy or transfer rights
- Record an authority and rights decision
Primary deliverable: Source authority-and-rights register.
03Map users to source access
Map user identities to permitted sources and identify access questions for the right owner.
Five practical steps
- Identify the helpdesk and specialist identities
- Map allowed and denied source pairs
- Trace the restricted engineering note
- Test a changed access state
- Route unclear rules to the identity owner
Primary deliverable: Identity–source permission matrix.
04Specify what the corpus needs
Specify document structure, metadata and refresh needs for a small searchable corpus.
Five practical steps
- List source formats and identifiers
- Record version and update fields
- Inspect the manual table structure
- Specify validation samples
- Document refresh and ownership needs
Primary deliverable: Corpus intake specification.
05Route a prebuild decision
Present a source or access dependency with options and a clear owner handoff.
Five practical steps
- State the unresolved source-use choice
- Separate source and identity decisions
- Compare feasible routes
- Record evidence and hold points
- Send each question to its owner
Primary deliverable: Prebuild decision-and-handoff record.
Prepare trustworthy, current source material
Choose permitted source routes, inspect extraction and passage design, validate the index, and test a failed update using the supplied case.
06Choose how a source is reached
Compare indexing and federation routes using rights, freshness and failure behavior.
Five practical steps
- Compare local indexing and query-in-place
- Check source-use rights
- Check identity propagation
- Compare freshness and outage behavior
- Record a bounded source route
Primary deliverable: Source-route comparison sheet.
07Check document extraction
Detect extraction errors that change the meaning of retrieved documents.
Five practical steps
- Inspect the original manual span
- Compare extracted heading and table
- Locate changed or missing meaning
- Quarantine defective passages
- Record a repair and retest
Primary deliverable: Document-extraction QA log.
08Design traceable passages
Design passages and metadata that retain meaning and trace back to their source.
Five practical steps
- Choose passage boundaries for two source types
- Retain source and section identifiers
- Keep version and span metadata
- Protect table labels and values
- Check traceability back to the original
Primary deliverable: Passage-and-metadata design sheet.
09Validate a searchable index
Validate index fields, filters and updates on an approved source set.
Five practical steps
- Inspect index fields and filters
- Run an exact-part query
- Identify a missing-field defect
- Apply an approved update or deletion
- Confirm obsolete hits disappear
Primary deliverable: Index validation record.
10Diagnose a failed source update
Detect stale search material and document a bounded recovery test.
Five practical steps
- Read the last successful bulletin sync
- Compare it with the supplied freshness rule
- Find affected queries and versions
- Record containment and owner
- Retest after the local repair
Primary deliverable: Freshness and recovery test card.
Retrieve evidence and assemble grounded answers
Compare judged lexical, semantic and hybrid search, test ranking trade-offs, and run a permission-aware answer path with exact source trace.
11Establish a judged search baseline
Build a judged lexical baseline with source and access filters.
Five practical steps
- Label the judged development questions
- Fix identity and expected sources
- Run lexical retrieval with filters
- Record exact-match success and paraphrase miss
- Keep holdouts separate
Primary deliverable: Lexical baseline results sheet.
12Compare semantic retrieval
Compare semantic results with a fixed judged baseline and access boundary.
Five practical steps
- Reuse the same judged cases and identities
- Run semantic retrieval
- Check a recovered paraphrase
- Inspect an irrelevant semantic hit
- Confirm denied material stays excluded
Primary deliverable: Semantic retrieval comparison.
13Choose a retrieval family
Select a retrieval approach from judged results rather than a default recipe.
Five practical steps
- Compare lexical, semantic and hybrid results
- Declare a choice and rejection condition
- Preserve access and source filters
- Open family holdouts once
- Record the selected retrieval path
Primary deliverable: Hybrid/family selection memo.
14Test one ranking change
Test a ranking change against relevance, access, delay and cost limits.
Five practical steps
- Set a within-family ranking change
- Run fixed development cases
- Compare source relevance and access
- Check p95 delay and cost
- Open the separate ranking holdouts once
Primary deliverable: Ranking-change experiment record.
15Build an answer path with source trace
Run a small permission-aware RAG path with claim-level source trace.
Five practical steps
- Carry identity into the request contract
- Filter retrieval and answer context
- Resolve citations under the same access rule
- Run four answer states
- Export source-span and version traces
Primary deliverable: Executable grounded-answer prototype.
Evaluate, protect and hand off the service
Test relevance, claim support, access changes and insufficient evidence separately, then hand measured results and open decisions to accountable owners.
16Build a regression case set
Create a versioned test set for relevance, answer support and failure cases.
Five practical steps
- Version questions and expected identities
- Add absent and stale cases
- Add conflict and citation-support cases
- Add revocation and denied-source cases
- Reproduce one defect on a sealed split
Primary deliverable: Versioned full evaluation-case set.
17Check each claim against its passage
Verify that every consequential answer claim is supported by its cited passage.
Five practical steps
- Split an answer into consequential claims
- Open each exact cited passage
- Check identity, version and span
- Mark supported and unsupported claims
- Record a reasoned adjudication
Primary deliverable: Claim–evidence adjudication ledger.
18Test access changes end to end
Test revocation and denied-source behavior across retrieval, answers and citations.
Five practical steps
- Test allowed and denied identities
- Revoke a source permission
- Inspect retrieval and answer context
- Check generated text and citation lookup
- Escalate and retest any leak
Primary deliverable: Access regression result set.
19Decide when evidence is insufficient
Choose safe answer, retry, insufficiency or escalation behavior from evidence.
Five practical steps
- Classify the available evidence
- Select supported answer or one permitted retry
- Use insufficiency for missing authorized support
- Escalate conflicting versions
- Record safe text and named owner
Primary deliverable: Insufficient-evidence decision protocol.
20Hand off measured results and feedback
Present measured release evidence and route feedback, defects and recovery work.
Five practical steps
- Bring together quality and access results
- Check freshness, delay and cost
- Reproduce a wrong-answer report
- Route a stale-sync alert
- Hand off defects and release decisions
Primary deliverable: Release-and-operating handoff packet.
Applied capstone
Resolve a contractor guidance answer with source evidence.
Use the course methods that fit the situation to make one supportable decision brief.
The situation
At fictional Pinehaven Design Cooperative, a contractor asks whether they may self-enroll a replacement laptop and what the helpdesk should do after a failed attempt. A plausible answer cites a current employee policy, while contractor guidance, a retired quick-start, restricted material and a limited vendor source create different evidence and access boundaries.
Your task
Prepare one annotated brief for the helpdesk service owner. Check each consequential claim against the current permitted passage, run the supplied offline access and citation checks, compare the unsafe candidate with a supportable path, and route unresolved policy, identity and service decisions to their owners.
The people behind MTF
Meet MTF faculty and the learner community.
Explore the professional backgrounds of MTF faculty and learn more about the international community studying with the Institute.
Enrollment
Enroll in Professional Certificate in RAG, Enterprise Search & Knowledge Systems
One-time course price: €10, including applicable taxes. Payment is processed securely by Stripe. No card details are stored on the MTF Institute website.
You will receive an email with access to the course. If you have any difficulties, please write to welcome@gtf.pt.
Questions and details
Frequently asked questions
Open the sections that matter to you, including delivery format, AI-supported practice and the evidence used to design the curriculum.
Who is this course for?
For early-career and transitioning search, AI, knowledge-systems, platform and application professionals who need a practical route from a defined information task to a tested, supportable search feature.
How does the course work?
The course is online and self-paced, with four modules, 20 applied lessons and one separate capstone. Work through supplied source cards, templates and a local offline lab; the displayed study time is up to one month.
What will I build and test?
You will prepare source and access records, inspect extraction and an index, compare judged retrieval methods, run a small permission-aware grounded-answer path, test claim support and access changes, and hand measured results to accountable owners.
How is AI used in the practical work?
The lessons use AI-assisted drafting and challenge where permitted. Learners verify source rights, user access, exact cited passages, versions, answer states and human decisions against the supplied case and local checks.
What evidence supports the curriculum?
The course draws on an MTF vacancy study, an independent article on current enterprise search practice and the public research archive. Read the vacancy study, trends article and Zenodo version 2 record.
What is the applied capstone?
Use the fictional Pinehaven contractor-guidance case to create one annotated decision brief. Verify what current permitted passages support, inspect the local offline trace and route decisions to the source, identity and service owners.
What certificate and access will I receive?
After successful enrollment, learners receive access to the MTF learning platform. Completing the course provides an MTF Institute professional course-completion certificate for Professional Certificate in RAG, Enterprise Search & Knowledge Systems.