How Local TCV works

What this tool does

A conventional academic CV is usually a long list of positions, publications, grants, teaching, service, and awards. A thematic CV asks a different question: what do your most relevant contributions show about your ability to perform the role proposed in this application? The assistant helps you move from the list to that explanation.

The tool looks for useful facts in material you provide, organizes them into evidence cards, and asks you to fill important gaps. It then helps identify the strongest material for your discipline and proposed role. Once you have checked the information, a small language model can prepare a working draft of one official TCV section at a time.

The intended result is a starting draft that is easier to review and improve—not an application that should be submitted without editing.

When to use it

Use the assistant when a funding opportunity asks for the Tri-Agency CV or another closely related narrative CV. It is especially useful when your existing CV contains plenty of outputs but says little about your particular role, the quality of the work, who benefited, or what changed because of it.

You can also use it early in application planning. Adding the project summary and your proposed role helps reveal which career examples are relevant and which claims still need supporting information. Starting early gives you time to confirm outcomes with collaborators, trainees, partners, or institutional records.

Do not assume every Canadian funding opportunity uses this format. The CV required can vary by agency, competition, and participant role. Confirm the required document in the current funding-opportunity instructions before beginning.

What a thematic CV is

The shared Tri-Agency CV is a narrative document developed by SSHRC, NSERC, and CIHR. Rather than judging a researcher mainly through publication counts, journal reputation, grant totals, or citation measures, it gives applicants space to explain a broader range of contributions and their importance.

The format contains three sections: a personal statement; the most significant contributions and experiences; and supervisory and mentorship activities. The English document has a five-page total limit and the French document a six-page total limit. These pages are shared across the three sections, so the tool’s word targets are planning suggestions rather than official section limits.

Contributions may include publications, but they can also include software, datasets, methods, creative work, partnerships, policy influence, professional practice, community activities, public engagement, research leadership, standards, services, and other meaningful work. What matters is explaining why the contribution is relevant and significant, what you did, and what evidence supports the claim.

How the process works

1. Provide application context

The same career achievement may be central to one application and peripheral to another. Providing the opportunity, project summary, and your proposed role gives the assistant a basis for selecting relevant information. Avoid including confidential information that is unnecessary for this purpose.

2. Gather possible evidence

You can upload an existing CV, answer guided questions, or paste source text. If you provide a document, the assistant separates long CVs into smaller subject-based chunks, extracts each one, and merges repeated facts. If a model returns incomplete JSON, the assistant attempts a constrained repair and extraction retry. Extraction is still only a first pass and can miss or misunderstand information.

3. Verify and enrich the information

You review each contribution before drafting. The assistant asks about your distinct role, why the work matters, what activities helped it reach relevant people, who engaged or used it, what changed, what funding or resources enabled, and how the experience relates to the application. Blank fields are prompts to investigate—not permission for the AI to guess.

4. Find the strongest material

The tool scores candidate contributions using signals appropriate to the selected discipline. You then choose no more than ten items, exclude weaker material, and drag or use arrow controls to put the evidence in the order the draft should follow. For example, policy or community uptake may be particularly informative in some social-science applications, while adoption of a method, tool, or technology may matter in an engineering context. These signals help organize your thinking; they do not reproduce an agency score or predict a funding decision.

5. Draft, edit, and export

The local model writes one section from the verified evidence. Each section has its own completeness gate, and saved sections contribute to a combined page-budget estimate. You can revise individual sections or export one combined draft with all three official headings, ready to paste into the required agency template. The final pagination and wording still need to be checked against the funding opportunity.

Choosing a starting pathway

CV plus guided evidence review — recommended for most researchers

This is usually the strongest starting route when you have a current PDF, Word, or plain-text CV. The local model gathers dates, titles, outputs, grants, roles, and other recorded facts. You then complete the same structured impact and mentorship questions used in the interview route. This hybrid matters because conventional CVs are useful factual records but often omit the activities, relationships, use, and outcomes needed for a narrative CV. Scanned image PDFs are not supported.

Build from guided questions — no CV required

This route starts with a structured interview instead of a document. It can work well for first-time applicants, people with non-linear careers, community and non-academic contributors, or anyone whose most important work is not represented by a publication list. The interview begins with the opportunity and proposed role, then prompts for research, community, professional, creative, technical, leadership, service, and mentorship experience. Add one evidence card for each contribution or connected body of work.

Paste selected source material

This flexible route accepts a biosketch, publication list, research summary, previous narrative CV, or working notes. It is useful when your source information comes from several places or when you want to provide a curated set rather than a complete CV. Extraction still requires verification, and you complete the same impact-pathway fields afterward.

What “evidence” means here

Evidence is information that allows a reviewer to understand and assess a claim. A publication citation proves that an output exists, but it does not by itself explain your role, the quality of the contribution, its relevance to the proposed work, or its influence. Those details often need to be added.

Useful evidence can be quantitative or qualitative. Examples include documented use by a partner, incorporation into policy or professional practice, adoption of software or data, improved access, a methodological advance, a trainee’s verified development, a community benefit, responsible research practices, or influence on later work. Numbers can provide context, but they should not substitute for explaining quality and impact.

The marker [EVIDENCE NEEDED] means an important idea is not yet supported by the information provided. Confirm the claim and add the evidence, rewrite it more cautiously, or remove it. Never replace the marker with a plausible guess.

Funding, activities, and a credible pathway to impact

Impact is not a synonym for publication, dissemination, popularity, or funding. A useful evidence chain separates five ideas: the contribution itself; the activity that helped it travel; the audience, participant, partner, or user reached; evidence of use or uptake; and an outcome or longer-term change. Not every contribution will have reached the final stage, and that is acceptable. Describe the strongest stage you can verify instead of implying a later one.

Activities can include co-design with communities or partners, advisory groups, policy briefings, workshops, training, public engagement, open data or software, plain-language or translated materials, demonstrations, pilots, licensing, implementation support, professional guidance, and ongoing feedback. These activities can strengthen a pathway to impact because they make work accessible, relevant, trusted, or usable. An activity belongs in the TCV only if it actually occurred; a planned future activity belongs in the research proposal or knowledge-mobilization plan unless the opportunity says otherwise.

Reach describes who encountered or participated in the work. Uptake describes a stronger action, such as testing, adapting, adopting, citing, incorporating, or reusing it. Outcomes are nearer-term changes in knowledge, capability, access, behaviour, practice, policy, process, or research direction. Longer-term impacts may be academic, health, environmental, economic, cultural, professional, community, policy, or societal. Use careful language such as “contributed to” when several causes were involved, and reserve causal wording for claims supported by appropriate evidence.

Funding is an input. A grant may be relevant because it demonstrates your role, supported trainees, enabled data collection, created access to equipment, sustained a partnership, funded engagement, or made implementation possible. The more useful sentence explains that connection: “The program supported partner workshops and trainee time, enabling the team to co-develop and test the resource.” A dollar amount, number of grants, or award success does not by itself demonstrate the quality or impact of the resulting work.

The guided fields use sentence starters as scaffolding. Replace every bracketed idea with your own verified detail. For example: “To make the findings usable by [audience], I [activity]. [User or partner] then [used or adapted the contribution], as shown by [source of evidence]. This contributed to [documented outcome].” If one link is unknown, leave it blank, seek confirmation from records or collaborators, or state the result more cautiously.

Local AI and privacy

Most online AI services send your text to computers operated by the service provider. This assistant instead downloads a smaller AI model and runs it inside your browser. Your CV content, extracted text, application context, and draft are not intentionally sent to an external model service or this website’s server.

“Local” does not mean that no internet connection is used. The page, fonts, software, and model files must be downloaded, and the organizations serving those files may receive ordinary technical information such as your IP address. The model can also occupy substantial browser storage. Clearing the site’s stored data can remove the cached files.

Local models offer a privacy advantage but are generally less capable than leading cloud models. They may overlook facts, lose context in long documents, produce awkward prose, or misunderstand a relationship. The evidence-review stage is therefore a central part of the approach, not an optional proofreading step.

The Settings panel can load WebLLM’s current catalogue of compatible instruction models. Smaller models generally download faster and work on more devices; larger models may draft more effectively but require substantially more storage and graphics memory. A model appearing in the catalogue does not guarantee that a particular device can run it.

The device check reads the browser’s WebGPU limits and, where available, a broad system-memory estimate to recommend a model size. Browsers do not expose exact free graphics memory, so the recommendation reduces—but cannot eliminate—out-of-memory failures. Qwen 2.5 and Phi 3.5 options are available alongside Llama models.

Saving work is optional. If you enable “Save my TCV workspace in this browser,” the source text, evidence, selections, and drafts are stored in IndexedDB on that device. A visible Clear button removes the stored copy. This information is not synchronized to another browser or computer.

Desktop version

Web Assistant version 0.1.0. The desktop app is released separately and may have a newer version number. The current desktop release adds adaptive agency positioning, SSHRC/NSERC/CIHR evidence review, and local-model refinement. See the desktop help or browse the Mac and Windows version archive.

If you prefer to install an app on your computer, download Local TCV Desktop for Mac or Windows. Unzip the file, double-click the installer, and follow the prompts. Your CV stays on your device.

Prefer not to install anything? Use the browser assistant instead (needs a recent Chrome, Edge, or Firefox with WebGPU).

SSHRC, NSERC, and CIHR planning modes

The assistant offers three planning lenses. SSHRC emphasizes interpretation, context, communities, and knowledge mobilization; NSERC emphasizes technical contribution, rigor, validation, and implementation; CIHR emphasizes health relevance, populations, partnerships, implementation, and outcomes.

Each mode asks agency-specific evidence questions and shows a fit check before drafting. These are practical framing aids informed by common application expectations, not official agency rules. Researchers working across fields can compare lenses and should always follow the current funding opportunity instructions.

What the tool cannot do

The assistant cannot determine whether a claim is true merely because it sounds credible. It does not know facts that are absent from the supplied material, and it cannot reliably distinguish every nuance of authorship, supervision, partnership, or impact. You are responsible for confirming all names, dates, quantities, roles, outcomes, and sensitive statements.

After drafting, the unsupported-details check compares names, multi-word entities, acronyms, and numbers against the verified evidence. It flags items that are not found verbatim. This is a conservative warning system: it can produce false alarms and cannot prove that unflagged prose is true.

The discipline ranking is editorial support, not an official agency assessment. It cannot know how a particular review committee will weigh your work, and it should not push every career into the same pattern. Applicants should use their knowledge of the project, field, communities, and evaluation criteria to override weak suggestions.

The tool also does not submit an application, guarantee correct formatting, or automatically satisfy opportunity-specific appendices. Requirements can change, so built-in guidance may become outdated.

Before submitting

Read the current funding-opportunity instructions and download the template provided for that competition. Confirm the required CV type, participant role, language, page limit, formatting, file format, appendix requirements, and submission method.

Then review the complete TCV as a single document. Check that it clearly supports your proposed role, stays within the shared page limit, is understandable without opening outside links, and does not repeat low-value information. Confirm that no unsupported markers remain and that any personal, lived-experience, career-interruption, Indigenous, community, or equity-related information is accurate, relevant, and included intentionally.

A trusted colleague or research-services professional can provide a valuable final review. Ask whether the document makes your contribution and relevance clear—not simply whether the prose sounds polished.

Official resources

Use these official pages as the authority when they differ from this documentation:

• Tri-Agency CV instructions
• Guidelines for reviewing the Tri-Agency CV
• Tri-Agency CV frequently asked questions
• SSHRC Tri-Agency CV overview
• NSERC Tri-Agency CV resources
• CIHR guidance on contributions and impacts

Documentation reviewed July 25, 2026.