# AI use cases for Kev (jaredpalmer/kev) Repo: [jaredpalmer/kev](https://github.com/jaredpalmer/kev) | Use cases: 3 | Page: https://dedupedai.com/use-cases/jaredpalmer-kev/ Kev is a family of small decision models (0.8B to 27B) built on Qwen3.5 and Qwen3.8 that answer yes/no (noul), multiple-choice (choice) and rating (score) questions about a piece of text in one request, with calibrated probabilities. Its server matches TypeSafe's System One API so the TypeSafe Python SDK works unchanged, and you can fine-tune on your own labels with a coding-agent skill on Modal. This file holds every use case of this repo. Each one starts with a "## " heading and ends with its URL. - Serve a local decision model that routes support tickets: https://dedupedai.com/use-cases/jaredpalmer-kev/serve-kev-for-ticket-routing/ - Point the TypeSafe SDK at your own Kev server: https://dedupedai.com/use-cases/jaredpalmer-kev/swap-jev-for-a-self-hosted-kev-server/ - Fine-tune Kev on your own support tickets with an agent skill: https://dedupedai.com/use-cases/jaredpalmer-kev/fine-tune-kev-on-your-labels/ ## How to serve a local decision model that routes support tickets with Kev Repo: [jaredpalmer/kev](https://github.com/jaredpalmer/kev) | Difficulty: Intermediate | Time: 45 minutes | Basis: From the README Run Kev-4B locally and send a ticket with department, escalate and frustration questions in one request. **Who it is for:** A support engineer who wants routing and escalation decisions without per-call API costs. **The problem:** Chat models give free text that needs parsing and their confidence is not calibrated. **The outcome:** Structured answers with calibrated probabilities, such as returns, a 0.93 escalate probability and a frustration score. ### Steps 1. Run `git clone https://github.com/jaredpalmer/kev.git && cd kev` 2. Install the server extras with `uv sync --extra serve` 3. Start the server: `uv run --extra serve python -m kev.serve --run jaredpalmer/kev-4b --port 8009` 4. POST a ticket with choice, noul and score questions to localhost:8009/v1/systemone as in the README curl example 5. Route tickets by the returned probabilities and thresholds you choose ### Prompt to paste into your agent ```text Set up jaredpalmer/kev with Kev-4B on port 8009, then write a script that sends each ticket in with a department choice, an escalate noul and a frustration score, and writes the answers and probabilities to a CSV. ``` ### Caveat Do not pip install Kev: that name on PyPI is an unrelated ORM. Install from source as shown. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/jaredpalmer-kev/serve-kev-for-ticket-routing/ ## How to point the TypeSafe SDK at your own Kev server with Kev Repo: [jaredpalmer/kev](https://github.com/jaredpalmer/kev) | Difficulty: Intermediate | Time: 30 minutes | Basis: From the README Keep your existing TypeSafe client code and switch its base URL to a local Kev server. **Who it is for:** A team already using Jev through TypeSafe that wants a self-hosted option. **The problem:** Hosted decision APIs cost per call and send your data to a third party. **The outcome:** The same client code calling Kev locally, with the API key set to local. ### Steps 1. Start a Kev server as in the quick start 2. Create a TypeSafeClient with api_key local and base_url http://127.0.0.1:8009 as in the README 3. Call client.system_one with your existing questions 4. Compare answers and probabilities with your hosted results on a sample 5. Pick a model size: Kev-4B to start, Kev-9B or Kev-27B for more accuracy ### Prompt to paste into your agent ```text Change my TypeSafe client code in to use a local jaredpalmer/kev server at 127.0.0.1:8009, run 50 sample inputs through both the hosted API and Kev, and report agreement and latency. ``` ### Caveat Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/jaredpalmer-kev/swap-jev-for-a-self-hosted-kev-server/ ## How to fine-tune Kev on your own support tickets with an agent skill with Kev Repo: [jaredpalmer/kev](https://github.com/jaredpalmer/kev) | Difficulty: Advanced | Time: half a day | Basis: From the README Install the kev-finetune skill and ask your agent to fine-tune Kev on your data. **Who it is for:** An ML-curious developer with labeled tickets or complaints whose zero-shot accuracy is too low. **The problem:** Generic models do not know your routing categories or escalation rules. **The outcome:** A fine-tuned Kev with a temperature fitted to your labels, served from your own endpoint. ### Steps 1. Run `npx skills add jaredpalmer/kev@kev-finetune` 2. Ask your agent to fine-tune Kev on your support tickets 3. Let the skill interview you and find the questions your code already asks 4. Train on Modal, check accuracy on held-out data and size your dataset first 5. Deploy the result as your own HTTPS endpoint that scales to zero ### Prompt to paste into your agent ```text Use the kev-finetune skill from jaredpalmer/kev to fine-tune Kev-4B on my labeled tickets in . Report accuracy before and after on a held-out split, and stop before deploying if the gain is within noise. ``` ### Caveat The README notes that with about 400 records the gain was inside the noise, so size your dataset before training. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/jaredpalmer-kev/fine-tune-kev-on-your-labels/