EverCV now tracks ML engineering work: W&B runs, HuggingFace publishes, LangSmith traces

EverCV hit 115 signal sources tonight.

The five new ones are different from everything else in the list. They're not CI/CD pipelines or project tracking systems — they're the tools that ML engineers and AI engineers actually live in, and they've been invisible to every CV tool that exists.

The problem

When a software engineer wants to update their resume, they have GitHub. Commits, PRs, releases — it's all there, timestamped, public.

When an ML engineer wants to update their resume, they have... what exactly? Their best work is often a W&B run that found the hyperparameter combination that pushed F1 from 0.81 to 0.87. Or a HuggingFace model deployment that their team used for six months. Or 200 Semgrep findings they methodically fixed over a quarter. None of that shows up in a commit diff in any meaningful way.

The result: ML engineers have the hardest time writing accurate CVs because their highest-leverage work is the least visible.

What's new

Weights & Biases — When you finish an experiment run (state: finished), EverCV sees it. "Completed W&B run: hyperparameter-sweep-v3 (entity/project)." The run name, the project, the timestamp — all captured. If you ran 50 experiments last quarter to find the right architecture, those 50 runs are 50 evidence points of ML engineering rigor.

Hugging Face — When you publish or significantly update a model, space, or dataset, EverCV sees it. "Updated HuggingFace model: username/bert-finetuned-ner." Models, Spaces, and Datasets — all three. If you maintain public ML artifacts, this finally gives you a clean record of when you shipped each version.

LangSmith — When an LLM evaluation chain completes successfully, EverCV sees it. LangSmith traces are the unit of work for LLM engineers the same way commits are for software engineers. If you're building production AI systems with LangChain or similar, this is the signal source that makes your evaluation discipline visible.

Semgrep — When a Semgrep finding is marked fixed, EverCV sees it. "Fixed Semgrep finding: python.django.security.injection.tainted-sql-string in auth/views.py:47." Resolving a static analysis finding is real security work. For DevSecOps engineers and security-minded backend engineers, this surfaces work that would otherwise be invisible.

Spacelift — When a Spacelift IaC stack apply finishes, EverCV sees it. "Spacelift applied: production-eks-cluster (run run-abc123)." For platform engineers who manage infrastructure through Spacelift rather than raw Terraform/OpenTofu commands, this is the deployment signal source they've been missing.

Why these five

The common thread: these are all tools where the output isn't a code commit.

An ML experiment produces a run, not a PR. A published model produces a model version, not a release tag. A fixed SAST finding produces a resolved ticket in the security scanner, not a diff. An IaC apply produces an infrastructure state change, not a deployment log visible in GitHub Actions.

Every other CV tool I know of tries to solve the resume problem by extracting from GitHub. That works great for software engineers whose primary output is code. It works badly for anyone whose primary output is something else: trained models, infrastructure state, security posture, LLM evaluations.

EverCV is now the only CV tool that captures all of these.

The full list

115 signal sources total. The ones that are different from anything else in the market:

| Category | Sources | |---|---| | ML / AI Engineering | Weights & Biases, Hugging Face (models + spaces + datasets), LangSmith | | Security | Snyk, Semgrep | | Platform Engineering | Kubernetes, Helm, AWS ECS, AWS CloudFormation, Octopus Deploy, Spacelift | | Data Engineering | dbt Cloud, Prefect | | Code Quality | SonarQube |

The rest are the standard engineering stack — GitHub, GitLab, Jira, Linear, CircleCI, PagerDuty, Datadog, Vercel, and 90+ more. Everything merges into a single timeline and gets rendered into resume bullets automatically.

How it works

EverCV connects to your tools via API keys you control. Once connected, it polls daily (Prosumer tier) or weekly (Free tier), extracts signal from each source, and generates or updates your resume.

For W&B: connect your entity, project, and API key. Every time a run finishes, it shows up in your CV timeline.

For HuggingFace: connect your username and token. Every model/space/dataset update shows up.

For LangSmith: connect your API key and project ID. Every successful evaluation trace shows up.

The extraction layer translates each raw signal into a resume bullet: what you did, when you did it, which tool, in active past-tense language that reads like a human wrote it.

EverCV is at evercv.io. Free tier is GitHub only. Prosumer ($15/mo) gets all 115 signal sources and daily updates.

If you're an ML engineer or AI engineer who's tried to explain a year of model experiments on a traditional resume, I'd like to know if this helps. hello@evercv.io