<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Tech | Lee-Ann Vidal Covas, PhD</title><link>https://leeannvc.com/en/tags/tech/</link><atom:link href="https://leeannvc.com/en/tags/tech/index.xml" rel="self" type="application/rss+xml"/><description>Tech</description><generator>HugoBlox Kit (https://hugoblox.com)</generator><language>en-us</language><image><url>https://leeannvc.com/media/logo_hu_408c0977b7e48a52.png</url><title>Tech</title><link>https://leeannvc.com/en/tags/tech/</link></image><item><title>AI Model Evaluation: Fact-Grounded Tests for Frontier Models</title><link>https://leeannvc.com/en/tech_projects/ai_evaluation/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://leeannvc.com/en/tech_projects/ai_evaluation/</guid><description>&lt;h2 id="overview"&gt;Overview&lt;/h2&gt;
&lt;p&gt;As an AI Training Expert at Micro1, I&amp;rsquo;ve designed knowledge-seeking evaluation tasks for frontier AI models. Each task asks whether a model can answer a question grounded in one indisputable, source-verifiable fact, and each is built to reveal where models fail. Project and client details are confidential, so this page describes the method rather than the content.&lt;/p&gt;
&lt;h2 id="my-role"&gt;My Role&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;Designed question-answer evaluation tasks, each anchored to a single, indisputable, source-verifiable fact&lt;/li&gt;
&lt;li&gt;Engineered and tested prompts against live model outputs over multiple trials, refining each question until it reliably exposed a genuine failure mode&lt;/li&gt;
&lt;li&gt;Vetted niche, specialized sources against strict accessibility and reliability criteria&lt;/li&gt;
&lt;li&gt;Wrote quote-supported justifications that tie every claim to primary-source text&lt;/li&gt;
&lt;/ul&gt;
&lt;h2 id="workflow--quality"&gt;Workflow &amp;amp; Quality&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;Every task went through a multi-stage peer quality review: initial review, independent verification, and sign-off&lt;/li&gt;
&lt;li&gt;Tasks were accepted only when they met shared evidentiary standards&lt;/li&gt;
&lt;/ul&gt;
&lt;!-- On hold (Lee-Ann is deciding whether to keep it):
## Why a Linguist
- Writing a question with exactly one correct reading is a problem of meaning and ambiguity. My linguistics training helps me spot where a question could be read two ways, and where a model's answer only looks right.
--&gt;</description></item><item><title>Cogito (now part of Verint): Speech Data Annotation for Machine Learning</title><link>https://leeannvc.com/en/tech_projects/cogito/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://leeannvc.com/en/tech_projects/cogito/</guid><description>&lt;h2 id="overview"&gt;Overview&lt;/h2&gt;
&lt;p&gt;Cogito (now part of Verint) builds AI that analyzes voice and conversational signals during contact-center calls to give agents real-time guidance on empathy, pacing, and engagement.&lt;/p&gt;
&lt;p&gt;I worked on annotation and quality assurance for speech and language datasets used in machine learning models, focusing on improving model performance and annotation consistency.&lt;/p&gt;
&lt;h2 id="my-role"&gt;My Role&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;Annotated speech data for emotion, engagement, and speech patterns&lt;/li&gt;
&lt;li&gt;Designed and refined annotation approaches across projects&lt;/li&gt;
&lt;li&gt;Conducted prompt engineering to improve model outputs&lt;/li&gt;
&lt;li&gt;Tested pre-trained language models and suggested calibration improvements&lt;/li&gt;
&lt;/ul&gt;
&lt;h2 id="data--workflows"&gt;Data &amp;amp; Workflows&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;Processed audio and text data for ML training pipelines&lt;/li&gt;
&lt;li&gt;Built and validated annotated datasets for internal and external clients&lt;/li&gt;
&lt;li&gt;Handled dynamic annotation requests across teams&lt;/li&gt;
&lt;li&gt;Contributed to workflow improvements and QA processes&lt;/li&gt;
&lt;/ul&gt;
&lt;h2 id="impact"&gt;Impact&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;Improved annotation consistency across datasets&lt;/li&gt;
&lt;li&gt;Contributed to higher-quality training data for ML models&lt;/li&gt;
&lt;li&gt;Provided insights that informed model behavior and UX decisions&lt;/li&gt;
&lt;/ul&gt;
&lt;h2 id="highlight-research-that-changed-a-product-decision"&gt;Highlight: Research That Changed a Product Decision&lt;/h2&gt;
&lt;p&gt;A proposed in-call guidance feature would have coached agents to stop saying &amp;ldquo;uh&amp;rdquo; and &amp;ldquo;um.&amp;rdquo; My manager suspected it was a mistake and asked for my help, since my research is on filled pauses. I explained the research and built a presentation showing why the feature would backfire: filled pauses help speakers hold the floor while they plan, signal that more is coming, and cue listeners to pay attention. Coaching them out would make agents sound more robotic and undercut the turn-taking and customer-experience behaviors the product already encouraged. My manager brought the case to the relevant stakeholders, and the feature was dropped.&lt;/p&gt;</description></item><item><title>Application Build System: Modular LaTeX &amp; AI Workflow Automation</title><link>https://leeannvc.com/en/tech_projects/build_system/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://leeannvc.com/en/tech_projects/build_system/</guid><description>&lt;h2 id="overview"&gt;Overview&lt;/h2&gt;
&lt;p&gt;Applying for roles in research, tech, and international organizations means producing many tailored documents that all have to stay accurate, consistent, and readable by applicant-tracking systems. I turned that work into a build system: documents are assembled from reusable, version-controlled modules, and AI agents take over the checks that used to be manual.&lt;/p&gt;
&lt;h2 id="stage-1-modular-latex-before-ai-tools"&gt;Stage 1: Modular LaTeX (before AI tools)&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;Built a cover-letter system on two axes, inspired by my partner&amp;rsquo;s simpler version: a &lt;strong&gt;mode&lt;/strong&gt; (academia, postdoc, tech, consulting, international organizations) selects the generic paragraphs, and a short &lt;strong&gt;per-application file&lt;/strong&gt; supplies the employer details&lt;/li&gt;
&lt;li&gt;Extended that per-application file to drive my research and teaching statements and writing samples, so one form controls every document in an application&lt;/li&gt;
&lt;li&gt;Any paragraph can be replaced for a single application; if the replacement is left empty, the generic version is used automatically&lt;/li&gt;
&lt;li&gt;Wrote optional-field macros so a blank field (team, department, job reference) disappears cleanly instead of leaving stray words&lt;/li&gt;
&lt;li&gt;Compiled from VS Code with a Makefile and latexmk, with no separate LaTeX editor; combined packets merge the letter, statements, and CV into one PDF&lt;/li&gt;
&lt;li&gt;A new tailored letter meant filling in one short file instead of rewriting a whole document&lt;/li&gt;
&lt;/ul&gt;
&lt;div class="text-center"&gt;
&lt;a
id="button-d950093277c08ca9755d3f002f5646d3"
href="https://github.com/leeannvidal/latex-application-kit"
target="_blank"
rel="noopener noreferrer"
class="inline-flex items-center gap-2 font-medium no-underline transition-all duration-300 ease-out transform-gpu focus:outline-none focus:ring-4 focus:ring-offset-2 focus:ring-offset-white dark:focus:ring-offset-zinc-900 disabled:opacity-50 disabled:cursor-not-allowed disabled:pointer-events-none bg-white dark:bg-zinc-900 border-2 border-primary-500 text-primary-600 dark:text-primary-400 hover:bg-primary-50 dark:hover:bg-primary-950/50 hover:border-primary-600 active:bg-primary-100 dark:active:bg-primary-950 shadow-md hover:shadow-lg hover:scale-105 active:scale-95 focus:ring-primary-500/50 px-4 py-2 text-base rounded-lg"
role="button"
aria-label="Template on GitHub, with a fictional example applicant"
&gt;
&lt;span class="flex-shrink-0"&gt;
&lt;svg class="w-4 h-4" xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"&gt;&lt;path fill="currentColor" d="M12 .297c-6.63 0-12 5.373-12 12c0 5.303 3.438 9.8 8.205 11.385c.6.113.82-.258.82-.577c0-.285-.01-1.04-.015-2.04c-3.338.724-4.042-1.61-4.042-1.61C4.422 18.07 3.633 17.7 3.633 17.7c-1.087-.744.084-.729.084-.729c1.205.084 1.838 1.236 1.838 1.236c1.07 1.835 2.809 1.305 3.495.998c.108-.776.417-1.305.76-1.605c-2.665-.3-5.466-1.332-5.466-5.93c0-1.31.465-2.38 1.235-3.22c-.135-.303-.54-1.523.105-3.176c0 0 1.005-.322 3.3 1.23c.96-.267 1.98-.399 3-.405c1.02.006 2.04.138 3 .405c2.28-1.552 3.285-1.23 3.285-1.23c.645 1.653.24 2.873.12 3.176c.765.84 1.23 1.91 1.23 3.22c0 4.61-2.805 5.625-5.475 5.92c.42.36.81 1.096.81 2.22c0 1.606-.015 2.896-.015 3.286c0 .315.21.69.825.57C20.565 22.092 24 17.592 24 12.297c0-6.627-5.373-12-12-12"/&gt;&lt;/svg&gt;
&lt;/span&gt;
&lt;span&gt;Template on GitHub, with a fictional example applicant&lt;/span&gt;
&lt;/a&gt;
&lt;/div&gt;
&lt;h2 id="stage-2-ai-agents--skills-claude-code"&gt;Stage 2: AI Agents &amp;amp; Skills (Claude Code)&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;ATS review skill:&lt;/strong&gt; a Python checker scans each built PDF for text-extraction hazards (ligatures, icon glyphs, footer text interleaving with the body, section headings); the agent then traces each finding to its source line and fixes it only on approval&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Repo-sync skill:&lt;/strong&gt; screens every change for credentials and personal records, groups the rest into logical commits, shows the plan for approval, and pushes without ever force-pushing&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Profile &amp;amp; drafting agents:&lt;/strong&gt; maintain a verified professional profile and writing voice, and suggest cover-letter edits grounded in the verified facts in that profile; I review every suggestion and decide what goes in&lt;/li&gt;
&lt;/ul&gt;
&lt;h2 id="design-principles"&gt;Design Principles&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;Write once, reuse everywhere:&lt;/strong&gt; each paragraph and field is written once and reused, so documents stay consistent&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Human in the loop:&lt;/strong&gt; agents propose, I approve&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Reproducible:&lt;/strong&gt; every document rebuilds from version-controlled source with one command&lt;/li&gt;
&lt;/ul&gt;</description></item><item><title>Communications On Hold: Business Website &amp; Audio Library</title><link>https://leeannvc.com/en/tech_projects/coh_website/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://leeannvc.com/en/tech_projects/coh_website/</guid><description>&lt;h2 id="overview"&gt;Overview&lt;/h2&gt;
&lt;p&gt;Communications On Hold produces custom voice content for businesses: on-hold messaging, IVR and call-routing prompts, and radio commercials. Customers choose voices and music by listening to samples, so the website is built around a large audio library that has to be easy to browse on a phone as well as on a desktop.&lt;/p&gt;
&lt;p&gt;
&lt;/p&gt;
&lt;h2 id="what-i-built"&gt;What I Built&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;Designed, built, and deployed the site with Hugo (a heavily customized HugoBlox theme), version-controlled on GitHub with continuous deployment through Netlify&lt;/li&gt;
&lt;li&gt;Wrote a custom audio-library component: a nested accordion of music, voice, and message samples that stays easy to use on mobile and desktop&lt;/li&gt;
&lt;li&gt;Wrote custom page blocks (page header, impact cards, contact card) and a custom CSS theme&lt;/li&gt;
&lt;li&gt;Configured the Netlify build with site-wide search (Pagefind) and 301 redirects from the company&amp;rsquo;s .net and .org domains to .com&lt;/li&gt;
&lt;/ul&gt;
&lt;h2 id="design-choices"&gt;Design Choices&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;Built around how customers actually use the site: listening to and comparing samples before they decide what&amp;rsquo;s best for their project&lt;/li&gt;
&lt;li&gt;A static site keeps pages fast and hosting simple&lt;/li&gt;
&lt;/ul&gt;
&lt;h2 id="ongoing"&gt;Ongoing&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;As the company&amp;rsquo;s Technology Consultant, I maintain the site and its domains alongside ongoing IT support&lt;/li&gt;
&lt;/ul&gt;
&lt;!-- Past-tense version, for if this role ends:
## Maintenance
- As the company's Technology Consultant, maintained the site and its domains alongside IT support
--&gt;</description></item></channel></rss>