Glossary

Technical terms, in plain language

Short explanations of specialized words used in Local TCV. These definitions explain the software; they do not replace current agency instructions.

Tri-Agency CV and research terms

Tri-Agency: Canada’s three federal research funding agencies: CIHR, NSERC, and SSHRC.
TCV (Tri-Agency CV): A short, narrative-style curriculum vitae used for certain Canadian research-funding applications. It asks applicants to explain selected contributions instead of submitting a complete career inventory.
CIHR: Canadian Institutes of Health Research. Its scope includes biomedical, clinical, health-services, implementation, population, and public-health research.
NSERC: Natural Sciences and Engineering Research Council of Canada. It supports natural-science and engineering research, training, partnerships, and related innovation.
SSHRC: Social Sciences and Humanities Research Council of Canada. It supports humanities, social-science, research-creation, partnership, and knowledge-mobilization work.
Agency lens: A planning view that helps notice evidence relevant to an agency’s broad research context. It is not an official score or eligibility decision.
Application context: The opportunity, project, audience, goals, and proposed role for which the TCV is being prepared.
Research contribution: Knowledge, understanding, a method, tool, dataset, creative work, intervention, relationship, program, or other meaningful advance produced through research.
Research-creation: Research in which artistic or creative practice is an important way of investigating a question and producing knowledge.
HQP (highly qualified personnel): Students, postdoctoral researchers, research staff, and other people who gain advanced research or technical skills through a project.
Indigenous research governance: Processes through which Indigenous Peoples exercise authority over research priorities, relationships, knowledge, data, interpretation, and benefits. Specific community protocols and distinctions must be respected.
Knowledge user: A person or organization positioned to use research, such as a community, policymaker, practitioner, patient, educator, business, or researcher.

Evidence, reach, and impact

Activity: Something done, such as holding a workshop, collecting data, supervising a trainee, or developing software.
Output: Something produced, such as an article, book, dataset, performance, policy brief, prototype, guideline, or piece of software.
Knowledge mobilization: Purposeful work that helps knowledge move between researchers and people who may use it. Examples include co-development, workshops, briefs, tools, exhibitions, training, and ongoing partnerships.
Reach: Who or how many people, organizations, places, or sectors encountered a contribution. Reach shows exposure, not necessarily use or benefit.
Uptake or adoption: Evidence that someone used, incorporated, implemented, cited, taught, licensed, deployed, or otherwise acted on a contribution.
Outcome: A documented near- or medium-term change connected to the work, such as changed practice, improved capability, a new policy, trainee progression, or measured health benefit.
Impact: A meaningful longer-term effect on knowledge, people, culture, policy, practice, health, the environment, the economy, or research capacity. Impact should be supported rather than assumed.
Significance: Why a contribution matters in its own context—for example, what gap it addressed, capability it enabled, perspective it changed, or need it met.
Evidence gap: An important detail that is missing or not yet supported, such as the applicant’s exact role, who used an output, or what changed afterward.
Validation: A check showing that a method, result, tool, intervention, or system performs as intended. The appropriate check varies by field.
Reproducibility: The ability to obtain consistent results again using sufficiently documented data, methods, code, or procedures.
Implementation science: The study of how evidence-based practices or interventions are adopted, adapted, delivered, sustained, and evaluated in real settings.
Patient-oriented research: Health research that meaningfully involves patients, caregivers, or communities in setting priorities, conducting the work, or applying results.
Sex- and gender-based analysis: Systematic consideration of how sex-related biology and gender-related social factors may shape research questions, methods, findings, and effects.

AI and local models

AI (artificial intelligence): Software that performs tasks associated with human reasoning or language. Local TCV uses AI to organize evidence and draft text, but its output can be incomplete or wrong.
LLM (large language model): An AI model trained to predict and generate language. It does not know whether every generated statement is true.
Local AI or on-device AI: AI processing that runs on the user’s own computer instead of sending the prompt and CV to a remote model service.
WebLLM: A software library that runs compatible language models inside a web browser using the device’s graphics capabilities.
WebGPU: A browser feature that gives web applications controlled access to graphics-processing hardware. The web version uses it to run the local model.
GPU: Graphics processing unit. It performs many calculations in parallel and can make AI models run faster.
Model: The downloaded AI program containing learned numerical patterns. Different models vary in size, speed, memory needs, and output quality.
Model weights: The large collection of numbers learned during model training. These files are downloaded before a local model can run.
Parameter: One adjustable number inside a model. Labels such as 1B or 3B roughly mean one billion or three billion parameters.
Quantization: A way to store model numbers with less precision so the model uses less disk space and memory, usually with some quality trade-off.
Prompt: The instructions and information given to an AI model for a task.
Token: A small piece of text processed by a language model. A token may be a word, part of a word, punctuation, or spacing.
Context window: The limited amount of text a model can consider during one request. Long CVs must be divided into smaller sections.
Chunking: Dividing a long document into smaller sections so each part fits within the model’s context window.
Hallucination: Plausible-sounding information generated by AI that is unsupported, inaccurate, or invented.
Inference: The calculation performed when a trained model reads a prompt and generates a response.
Ollama: A separate application that downloads and runs local AI models on Mac, Windows, or Linux. Local TCV Desktop communicates with it only through the local computer.
GGUF: A common file format for locally runnable language models, especially models used with llama.cpp-compatible software.
llama.cpp: Open-source software for running compatible language models locally, often from GGUF files. The name describes the software family, not only Meta’s Llama models.

Privacy, networks, and storage

Localhost: A network address that points back to the same computer. Local TCV Desktop uses 127.0.0.1, so its interface and Python service communicate on the local machine.
API (application programming interface): A defined way for software components to exchange requests and responses. The desktop browser interface uses a localhost API to ask the local model for results.
Cloud model: An AI model running on another organization’s computers. Using one requires sending input over the internet.
Browser cache: Files saved by the browser to make later visits faster. Local model files can be large and may be stored separately from ordinary page files.
Service worker: A browser component that can cache the public application files and help the site work after an initial successful load.
IndexedDB: Private storage managed by the browser for a particular website. Local TCV uses it only when the user enables workspace saving.
Cookie: A small value a website stores in the browser. Google Analytics may use analytics cookies to distinguish visits; CV text is not intentionally placed in them.
Metadata: Information about a request or file rather than its main content—for example, time, IP address, browser type, filename, or file size.
IP address: A network address used to route internet traffic. Websites and download providers ordinarily receive it when a device connects.
HTTPS: An encrypted connection between a browser and a website. It protects data in transit and is required for many modern browser features.
Google Analytics: A service used to measure page visits and interactions. Local TCV does not intentionally send CV text, extracted evidence, application context, or drafts to it.

Documents, installers, and software

PDF: Portable Document Format. Text-based PDFs can be read by Local TCV; image-only scanned PDFs require OCR first.
DOCX: The modern Microsoft Word document format. Local TCV reads its text locally in the browser.
OCR (optical character recognition): Software that converts pictures of printed or handwritten text into selectable text. OCR is not currently built into Local TCV.
JSON: A structured text format software uses to represent fields and values. Local TCV uses it internally to organize extracted evidence.
Python: A programming language used by Local TCV Desktop’s local service. Native installers bundle it, so end users do not need to install or learn Python.
Virtual environment or venv: A private folder containing Python packages for one application. It prevents one Python project from changing another. Native Local TCV installers make this unnecessary for end users.
Dependency: A separate software component an application needs in order to work.
PyInstaller: A packaging tool that bundles a Python application, its compatible Python runtime, and dependencies into a desktop application.
DMG: A Mac disk-image download commonly used to distribute an application, often by dragging the app into Applications.
PKG: A Mac installer package that guides installation into the Applications folder.
EXE: A Windows executable file. Local TCV’s Windows setup wizard is distributed as an EXE.
Code signing: A cryptographic publisher signature that helps the operating system verify who produced an application and whether it changed afterward.
Notarization: Apple’s automated security review for signed Mac software. A notarized release produces fewer security warnings for users.
Open source: Software whose source code is available under a licence that permits specified use, inspection, modification, and redistribution.

Accessibility

Accessibility: Designing information and interaction so people with diverse visual, hearing, motor, cognitive, and technological needs can use it.
ARIA: Accessible Rich Internet Applications. ARIA attributes give assistive technologies extra information about labels, states, relationships, and live updates when ordinary HTML is not enough.
Assistive technology: Hardware or software that helps a person use digital content, such as a screen reader, voice control, switch device, magnifier, or refreshable braille display.
Screen reader: Software that reads interface text and structure aloud or sends it to a braille display.
Keyboard focus: The control that currently receives keyboard input. A visible focus indicator shows keyboard users where they are.
Live region: An area marked so a screen reader can announce status changes, such as model-loading or extraction progress, without moving keyboard focus.
WCAG 2.2 AA: Widely used web accessibility guidelines. “AA” is the middle conformance level and covers requirements such as keyboard use, contrast, labels, focus, and understandable interaction.