Christian F. Rojas
ENES

Applied AI leadership, built on how people actually learn.

I lead technical documentation and put AI to work in it: an agent chosen by evaluation, traceable change methods, quality governance and a career path that treats AI as a skill. Some of this work is in production and some is still a proof of concept or a proposal; each case states its status and what it has not proven yet.

Christian F. Rojas, seated in front of a lit bookshelf, wearing a green checked shirt and smiling at the camera.
Focus
AI enablement · evaluation · knowledge systems
Sector
enterprise cybersecurity (digital fraud prevention)
Team
Leads a team of two writers
Background
Master's in neuroeducation
Based
Colombia · remote

Selected work

Five cases, each with its limit in plain sight.

Every case separates what I decided from what an AI assistant executed under my rules, and states what it does not prove yet.

115 traced changes across 3 releases · 3 agent architectures evaluated · 97% Recall@10 · 62 → 26 questions

Case 01

A documentation agent, chosen by measurement across three architectures

I wanted an agent that answers from our versioned technical manuals and always cites its source. I had three architectures built and evaluated, and chose with data, not opinion.

System diagram: three architectures for the documentation agent, each with its key figureThree architectures built in sequence. The first, retrieval-augmented generation (RAG), uses a retrieval index and a language model: LLM-judge score 4.46 of 5 and 12 of 17 answerable questions grounded. The second puts every manual of one version in the model’s context: 4.85 of 5 and 15 of 17. Both were scored on the same 26 golden questions. The third, highlighted, uses lexical retrieval with its own BM25 index and sub-agents, with no vector database: Recall@10 of 97%, Recall@5 of 91% and MRR of 0.79 on 34 paraphrased questions.1 · RAGretrieval indexlanguage model2 · full contextall manuals of one versionin the model’s context3 · BM25 + sub-agentsno vector databaseplan · select · answersame 26 golden questions · LLM judge34 paraphrased questionsjudge 4.46 / 512/17 groundedjudge 4.85 / 515/17 groundedRecall@10 97%Recall@5 91% · MRR 0.79System diagram: three architectures for the documentation agent, each with its key figureThree architectures built in sequence. The first, retrieval-augmented generation (RAG), uses a retrieval index and a language model: LLM-judge score 4.46 of 5 and 12 of 17 answerable questions grounded. The second puts every manual of one version in the model’s context: 4.85 of 5 and 15 of 17. Both were scored on the same 26 golden questions. The third, highlighted, uses lexical retrieval with its own BM25 index and sub-agents, with no vector database: Recall@10 of 97%, Recall@5 of 91% and MRR of 0.79 on 34 paraphrased questions.1 · RAGretrieval indexlanguage model2 · full contextall manuals of one versionin the model’s context3 · BM25 + sub-agentsno vector databaseplan · select · answerjudge 4.46 / 512/17 groundedjudge 4.85 / 515/17 groundedRecall@10 97%MRR 0.79
4.46 → 4.85
LLM-judge score (1–5), RAG against full context, on 26 golden questions
12/17 → 15/17
answerable questions with a grounded answer, RAG against full context
97%
Recall@10 for the third architecture on 34 paraphrased questions in Spanish and English

04 decisions of my own · Proof of concept · not in production

Read the case: A documentation agent, chosen by measurement across three architectures

Case 02

A change-plan method for documentation, in production across three releases

Change requests reached us as emails, tickets and version diffs thousands of lines long. I turned them into a versioned, traceable method and applied it to three product releases.

System diagram: traceability and escalation in the change-plan methodEvery item is traced in both directions along the chain verbatim quote, REQ, CH, P. When sources disagree, the contradiction goes to a contradiction log, highlighted, and is escalated, not settled on my own criteria. The method is in production: 3 releases, 115 changes traced.verbatim quoteREQCHPcontradiction logescalatednot settled by metraced both wayssources disagreein production · 3 releases · 115 changes tracedSystem diagram: traceability and escalation in the change-plan methodEvery item is traced in both directions along the chain verbatim quote, REQ, CH, P. When sources disagree, the contradiction goes to a contradiction log, highlighted, and is escalated, not settled on my own criteria. The method is in production: 3 releases, 115 changes traced.verbatim quoteREQCHPcontradiction logescalatednot settled by metraced both wayssources disagree3 releases · 115 changes
115
changes traced across 3 releases (43 + 56 + 16)
54
questions for subject-matter experts
27 · 14
requests traced and reviewer comments incorporated

04 decisions of my own · In production · 3 releases

Read the case: A change-plan method for documentation, in production across three releases

Case 03

Quality governance for AI-assisted documentation

My team started using AI to draft, translate and review, and nobody logged the errors it introduced. I set up checkpoints, an error log and two roles so those errors change the prompts.

System diagram: the quality loop for AI-assisted documentationAn AI draft, produced from prompts, passes 3 checkpoints, each with a checklist, before the approved output. The TW Reviewer applies the checks. The checkpoints record what they catch in an error log. The log feeds the TW Prompt Owner, highlighted, who adjusts the prompts for the next draft and closes the loop. The system is in production.AI draftfrom promptsTW Reviewer3 checkpointsa checklist eachapproved outputerror logwhat checks catchTW Prompt Ownercloses the loopappliesrecordsfeedsadjusts promptsin productionSystem diagram: the quality loop for AI-assisted documentationAn AI draft, produced from prompts, passes 3 checkpoints, each with a checklist, before the approved output. The TW Reviewer applies the checks. The checkpoints record what they catch in an error log. The log feeds the TW Prompt Owner, highlighted, who adjusts the prompts for the next draft and closes the loop. The system is in production.AI draftfrom prompts3 checkpointsapplied by TW Reviewererror logwhat checks catchapproved outputTW Prompt Ownercloses the loopfeedsadjusts promptsin production
3
checkpoints, one per stage: draft, translation, pre-export review
2
new roles: TW Reviewer and TW Prompt Owner

04 decisions of my own · In production · the team's formal AI-quality record

Read the case: Quality governance for AI-assisted documentation

Case 04

A 360° evaluation portal, with its scoring model corrected

HR's 360° evaluation lived in a spreadsheet. I directed three architecture generations and an analysis that cut one questionnaire from 62 to 26 questions and corrected how people are scored.

System diagram: how the career portal was diagnosed and reframedThe HR instrument is a spreadsheet. The proof of concept migrated only part of it. The diagnosis, highlighted, found the role-to-level matrix and the weights missing, and 62 questions against a design of 26. The project was reframed as a recovery of work HR had already done; the corrected model runs in the third generation, in production.HR instrumentspreadsheetproof of conceptpartial migrationdiagnosisno matrix or weights62 vs 26 questionsrecoveryof work HR already didGen 3 · in productionSystem diagram: how the career portal was diagnosed and reframedThe HR instrument is a spreadsheet. The proof of concept migrated only part of it. The diagnosis, highlighted, found the role-to-level matrix and the weights missing, and 62 questions against a design of 26. The project was reframed as a recovery of work HR had already done; the corrected model runs in the third generation, in production.HR instrumentspreadsheetproof of conceptpartial migrationdiagnosisno matrix or weights62 vs 26 questionsrecoveryof work HR already didGen 3 · in production
62 → 26
questions for one role once the role-to-level matrix was recovered (−58%)
~50%
chance that a person at the edge of a band changes band, which led to a near-threshold alert
41 · 7
findings fixed in two adversarial verification rounds, 7 of them blocking

04 decisions of my own · In production · Gen 3, self-contained

Read the case: A 360° evaluation portal, with its scoring model corrected

Case 05

A career framework that makes AI an evaluable competency

My documentation team had no explicit career path. I designed a four-level framework, now approved by HR, in which working well with AI is assessed like any other skill.

System diagram: the four-level documentation career framework and its AI competencyFour levels rise as a staircase: Junior Technical Writer, Intermediate Technical Writer, Senior Technical Writer and Documentation Manager, each with an expected profile of functions, hard skills and soft skills. The AI competency, highlighted, runs through the levels: one function and one hard skill, assessed like any other skill. The AI governance function goes from Basic at Junior to Advanced at Senior and Manager.Junior TWexpected profileIntermediate TWexpected profileSenior TWexpected profileDoc. Managerexpected profileAI competency1 function + 1 hard skillassessed like any other skillprofile = functions · hard · soft skillsBasicAdvancedAdvancedSystem diagram: the four-level documentation career framework and its AI competencyFour levels rise as a staircase: Junior Technical Writer, Intermediate Technical Writer, Senior Technical Writer and Documentation Manager, each with an expected profile of functions, hard skills and soft skills. The AI competency, highlighted, runs through the levels: one function and one hard skill, assessed like any other skill. The AI governance function goes from Basic at Junior to Advanced at Senior and Manager.Junior TWexpected profileIntermediate TWexpected profileSenior TWexpected profileDoc. Managerexpected profileAI competency1 function + 1 hard skillassessed like any other skillfunction: Basic → Advancedprofile = functions · hard · soft skills
4
levels, from Junior Technical Writer to Documentation Manager
≤ 12
active functions per level, so each profile stays focused
2
AI competencies, one function and one hard skill

05 decisions of my own · Approved · HR is implementing it

Read the case: A career framework that makes AI an evaluable competency

Shorter cases

Two shorter cases.

Smaller in scope, in production or in review, each with what it has not measured yet.

Status: In production

A documentation feedback system

Client-facing teams report documentation errors through one form, and a monthly report closes the loop after approval.

9
KPIs defined, from volume by type to backlog older than 30 days
< 1 min
to fill in a report

Assisted: An AI assistant built the static form and documented how it works, under my rules.

Status: In review with client-facing teams

One manual per situation, not per format

I directed a cut of about 65% to a legacy manual without losing procedural content. For the product's new version, I reworked my initial three-manual proposal into one manual organized by the reader's situation.

~65%
shorter legacy manual, against a target of 30–50%
3 → 1
proposed manuals, from one per format to one per situation
0
broken internal links in the new manual

Assisted: An AI assistant did the text reduction, the cross-check against sources and the final document, index and links under those rules.

How I work

Five rules I hold myself to.

Each one comes from decisions I made in real work.

How the work is split

I decide and direct; AI executes under my rules.

Christian F. Rojas leaning on the back of a chair, smiling.

  1. Ask before executing

    I ask before I act, and I made it a rule of my own. In a documentation audit I escalated 7 decisions to product leadership instead of settling them by assumption. In my change-plan method, a contradiction between sources is logged and escalated, not resolved on my own judgment.

  2. Measure abstention, not only accuracy

    A system that answers confidently when it should not is the riskier failure. My documentation agent's golden set included questions with no answer in the documentation, about another product and about another version. Both versions handled the trick questions 9 of 9.

  3. Think about the cost of reversal

    I weigh a decision by what it would cost to undo later. In the 360° portal, adopting the corrected model early costs almost nothing; changing it once the whole organization has been evaluated means telling everyone the model changed.

  4. Protect data by design, not by omission

    Third-party conversion tools were ruled out to keep information secure, so I had the converter built with no external dependencies; an AI assistant built it under my rules. The rule was simple: content does not leave the team.

  5. Stop the automation when you lose control

    In the documentation agent I chose a fixed flow over a tool-using agent, because the agent could not be forced to call its tool. Automation is worth keeping only while I can still verify what it does.

Lab

Proofs of concept, proposals and prototypes.

Two groups: what I have directed or proposed at work, then prototypes I design on my own.

01 · At work

At work: proofs of concept and proposals.

Work inside the company that is not in production. Each one says whether it is a proof of concept or a proposal.

Proof of conceptProcess proposal not implemented

A request system with service levels

A Kanban board for documentation requests, with SLIs, SLOs in three scenarios and SLAs per internal client.

What it includes

  • A request view for product leaders and a five-state Kanban for writers
  • 4 SLIs and P90 SLOs in three scenarios: conservative, standard and ambitious
  • SLAs per internal client, from product leaders to client-facing priorities

Assisted: An AI assistant built the board under my rules.

ProposalProposal submitted · ~42% reduction, estimated

Rebuilding an operations manual, with one house style

Product C's operations manual had nearly twice the words of a comparable manual and read like another company's. I proposed splitting it by audience and adopting one house style across products.

What it includes

  • A style comparison between two products' manuals: headings, procedure labels, how the reader is addressed and note density
  • Volume measured by chapter, with four redundancy patterns diagnosed
  • A two-document plan, Configuration and Operation, sharing one single-source preamble

Assisted: An AI assistant measured the volume, diagnosed the redundancy and drafted the tables of contents.

ProposalDelivered · decisions under review by product leadership

An audit of published documentation

Nobody knew which version of each document was official. I had the repositories mapped and escalated seven decisions to product leadership instead of guessing.

What it includes

  • A map of one product's documentation across its repositories
  • Source documents compared with what was published, with the gaps documented
  • A table for product leadership with 7 decisions escalated, each with its context and question

Assisted: An AI assistant inventoried the repositories, compared the versions and drafted the emails.

ProposalSubmitted to HR

Defining the documentation role and how it is evaluated

HR asked me to describe the documentation leadership role and how it should be evaluated. I answered with ranked responsibilities and an evaluation design built on four fairness conditions.

What it includes

  • A role description with 6 final accountabilities
  • Responsibilities ranked by strategic importance, not by time spent
  • Mirrored indicators of success and difficulty
  • An evaluation design with 4 fairness conditions

Assisted: An AI assistant drafted the text in English and then in Spanish, with my corrections.

02 · Personal lab

Prototypes, not products.

Six generic prototypes, unrelated to any employer. I designed and orchestrated each one to test an idea before it reaches a real team.

Prototype

Micro-learning Content Factory

Topic in, broadcast-ready video script out. I designed the pipeline so the instructional-design expertise lives inside it.

  • A script schema that enforces Gagné's Nine Events of Instruction
  • An intake web form where each field explains why it matters pedagogically

What I learned

Specialized sub-agents beat one mega-prompt. And making a private automation public and real-time (webhooks instead of forms, a database instead of an inbox) is its own discipline.

Try the live demo →

Prototype

Feedback Architect

Every learner comment is read on arrival and returned as sentiment, technical gaps and risk level.

  • A form with immediate persistence
  • An n8n "Silent Analyst": a single-purpose Gemini call

What I learned

A single-purpose judge prompt that returns one schema is more reliable at scale than a general summarizer that is also asked to classify. I design for what the admin needs to see now, not for what data we happen to have.

Prototype

CyberCoach, an adaptive assessor

A security assessment that tests readiness through live scenarios, then acts on the result.

  • Agent A: profiles the learner's role, generates situational scenarios live and scores competencies per topic
  • Agent B: when the conversation closes, writes a personalized follow-up email and curates resources for the gaps

What I learned

Splitting "assess" and "act on the assessment" into two agents simplified both prompts. The hard part was sequencing two asynchronous agents around one session ID.

Try the live demo →

Prototype

Post-Learning Intelligence

One training dataset, four stakeholder-specific reports, generated in a single run.

  • ROI calculated from KPI deltas
  • Four reports in one run: executive, ROI and finance, pedagogical audit, and operations alert

What I learned

Forcing reliable section boundaries in one LLM response took more than plain delimiters, which got swallowed. A marker that is not natural language solved it.

Try the live demo →

Prototype

Talent Mobility Orchestrator

A two-way conversational copilot: it fills a role from the workforce, or finds growth paths for one person.

  • An n8n AI agent that calls data tools on demand: profiles, DISC and open positions
  • Gemini scoring four dimensions: transferable skills, behavioral fit, performance trajectory and development investment

What I learned

When an agent needs precise data, build the precision into the data layer (RPC), not into the prompt.

Try the live demo →

Prototype

Orbit Docs, a RAG system

Answers from long product docs, with the source document and the exact passage on screen. Questions outside the docs get "not covered".

  • Gemini embeddings and a Pinecone index
  • An n8n agent that uses Pinecone as a tool

What I learned

Citation discipline mattered more than retrieval quality. An agent that finds the right passage but does not show which one is no more trustworthy than one that hallucinates.

Try the live demo →

Prototype

Event platform with privacy by design

A web platform for a private event, with its visual design adapted to that event: guests sign in with their phone, confirm attendance by family, and use dedicated pages on the day.

  • Sign-in by phone and attendance confirmation per family; each guest sees only their own family's data
  • Day-of pages: QR check-in with a seating plan, a live photo album, song requests and an organizers' panel

What I learned

Privacy has to be part of the build, not a review at the end: when the publish step itself refuses to ship private data, nobody has to remember to check.

Prototype

Interactive activities for an employability course

As a volunteer teacher at FIMLM, a nonprofit foundation, I taught a job-search strategies course every Friday in September and October 2026, and designed eight interactive activities for its virtual classes.

  • Scenario and story activities: following a résumé from posting to recruiter, a job seeker's branching path, and family advice to clear up before an interview
  • Game formats for the whole group: a treasure-hunt expedition, a quiz where every right answer moves the group up the ladder to employment, and hire-or-reject decisions

What I learned

Active retrieval beats explanation: when the group has to decide, answer in the chat and see the consequence, the content of the module comes back on its own.

About

An applied-AI leader, trained in how people learn.

I am Senior Manager, Technical Documentation in enterprise cybersecurity (digital fraud prevention). I own documentation for a line of fraud-prevention products and lead a team of two technical writers. My master's is in neuroeducation, so I start every AI system from how people learn and decide.

Degree

Máster en Neuropsicología, Inteligencias Múltiples y Mindfulness

Universidad Nebrija · IMF Smart Education · 2021–2024

Master's degree in education (neuroeducation) · 60 ECTS · 1,500 h · GPA 9.25

Selected certifications

  • Agentic AIDeepLearning.AI · 2026
  • A2A: The Agent2Agent ProtocolDeepLearning.AI · 2026
  • LLMOpsDeepLearning.AI · 2026
  • Governing AI AgentsDeepLearning.AI · 2026
  • Introduction to Model Context ProtocolAnthropic · 2026
  • Claude Code in ActionAnthropic · 2026
  • AI Fluency: Framework & FoundationsAnthropic · 2026
  • AI Fluency for EducatorsAnthropic · 2026
  • Essential AI Governance: Bridging Responsible AI, Compliance and RegulationLinkedIn Learning · 2026
  • Artificial Intelligence and Business StrategyLinkedIn Learning · 2026
  • Artificial Intelligence Micro-Certification (AIC)™Product School · 2026
  • Generative AI for Learning and Development Professional CertificateLinkedIn Learning · 2026
  • Prompt Engineering SpecializationVanderbilt University · Coursera · 2024
  • LXD SpecializationCoursera · 2024
  • n8n Course Level 1 · Automate Your Work with n8nn8n · LinkedIn Learning · 2025–2026
  • +5 short courses on AI for workforce learning, upskilling and career developmentLinkedIn Learning · 2026

Tools I work with

Claude Code · n8n · Power Automate · Airtable · Supabase · Postgres · Pinecone · Figma · Jira · GitHub

Contact

Building an applied AI function? Let’s talk.

Email is the most direct way to reach me. To learn more about my role or to request my CV, write to me there.

Based
Colombia · remote
Christian F. Rojas