DSML1: Setup Updates

.devcontainer/Dockerfile

@@ -0,0 +1,9 @@
+FROM mcr.microsoft.com/devcontainers/python:3.10
+
+COPY zscaler-bundle.pem /tmp/zscaler.pem
+RUN if [ -s /tmp/zscaler.pem ]; then \
+      cp /tmp/zscaler.pem /usr/local/share/ca-certificates/zscaler.crt && \
+      update-ca-certificates; \
+    fi && rm -f /tmp/zscaler.pem
+
+RUN find /etc/apt /usr/share/keyrings -name "*yarn*" -delete 2>/dev/null || true

.devcontainer/devcontainer-lock.json

@@ -0,0 +1,9 @@
+{
+  "features": {
+    "ghcr.io/devcontainers/features/docker-in-docker:3": {
+      "version": "3.0.1",
+      "resolved": "ghcr.io/devcontainers/features/docker-in-docker@sha256:ca2508495b01ba29eba93e8153772a2daa65eaa86471cc6863fe2a3d21933df9",
+      "integrity": "sha256:ca2508495b01ba29eba93e8153772a2daa65eaa86471cc6863fe2a3d21933df9"
+    }
+  }
+}

.devcontainer/devcontainer.json

@@ -1,7 +1,7 @@
 {
 	"name": "MLE Interview",
-	// Or use a Dockerfile or Docker Compose file. More info: https://containers.dev/guide/dockerfile
-	"image": "mcr.microsoft.com/devcontainers/python:1-3.10",
+	"build": { "dockerfile": "Dockerfile" },
+	"initializeCommand": "cp ~/.ssl/zscaler-bundle.pem .devcontainer/zscaler-bundle.pem 2>/dev/null || touch .devcontainer/zscaler-bundle.pem",
 	"customizations": {
 		"vscode": {
 			"settings": {
@@ -14,7 +14,7 @@
 						"source.organizeImports.ruff": true,
 						"source.organizeImports": true
 					}
-				},
+				}
 			},
 			"extensions": [
 				"charliermarsh.ruff",
@@ -29,7 +29,10 @@
 		}
 	},
 	"features": {
-		"ghcr.io/devcontainers/features/docker-in-docker": {}
+		"ghcr.io/devcontainers/features/docker-in-docker:3": {
+			"moby": false,
+			"disableIp6tables": true
+		}
 	},
 	"postCreateCommand": "make setup"
-}
+}

.gitignore

@@ -1,3 +1,6 @@
+# Zscaler cert (machine-specific, copied by initializeCommand)
+.devcontainer/zscaler-bundle.pem
+
 # Byte-compiled / optimized / DLL files
 __pycache__/
 *.py[cod]

INTERVIEW.md

@@ -1,28 +1,121 @@
-
 ## Coding Challenge
 
-You can refer to Google or relevant documentation to guide you in addressing the issues below. 
-You are not allowed to use ChatGPT, Copilot, or any similar tools for assistance. 
+You are working on a user interest recommendation system. It consists of:
+
+- A **TensorFlow model** trained on user content interactions, served via TF Serving (Docker)
+- A **Flask API** that fetches user features, calls the model, and returns ranked interests
+- A **feature store** backed by CSV files
+
+There are **7 tasks** to complete. Tasks 1–6 each have a failing test or a clear broken behaviour that tells you what to fix. Task 7 is an open-ended feature implementation. You can refer to Google or relevant documentation — no AI-assisted coding tools.
+
+---
+
+### Terminal setup (do this before task 4)
+
+Once the model is trained and served, you'll need **two terminals** running simultaneously inside the container:
+
+| Terminal | Command | Purpose |
+|---|---|---|
+| 1 | `make serve-model` | TF Serving on port 8501 |
+| 2 | `make serve-api` | Flask API on port 5005 |
+
+Keep both running while working on tasks 4–7.
+
+---
+
+### Tasks
+
+**1. Fix model training**
+
+Run `make test`. The `TestModel::test_model_build` test fails.
+
+Find and fix the bug in the model code, then verify:
+```
+make test  # TestModel::test_model_build must pass
+```
+
+Once passing, train the model and start the model server (leave it running):
+```
+make train
+make serve-model
+```
+
+---
+
+**2. Fix the API startup**
+
+Run `make serve-api`. The Flask application fails to start.
+
+Find and fix the bug, then verify the server starts without errors:
+```
+make serve-api
+```
+
+Leave the server running in its terminal.
+
+---
+
+**3. Fix the probability scores**
+
+The model's serving function returns raw logit scores, not probabilities. These values are not bounded between 0 and 1 and cannot be meaningfully compared or filtered.
+
+Find where the issue originates and fix it so that the scores returned represent proper probability values. There are two valid approaches — both are acceptable.
+
+Verify:
+```
+make test  # TestModel::test_probability_scores must pass
+```
+
+---
+
+**4. Fix the interests count**
+
+With both servers running, run `make test`. This test fails:
+```
+TestInterestsAPI::test_basic_response — AssertionError: Incorrect number of interests
+```
+
+Fix the API so it returns the correct number of interests. Verify:
+```
+make test  # test_basic_response must pass (interests count assertion)
+```
+
+---
+
+**5. Fix response time**
+
+With both servers running, run `make test`. This test fails:
+```
+TestInterestsAPI::test_basic_response — AssertionError: Request greater than 1 second
+```
+
+Find the performance bottleneck and fix it. Verify:
+```
+make test  # test_basic_response must pass (timing assertion)
+```
+
+---
+
+**6. Add probability to the response**
+
+With both servers running, run `make test`. This test fails:
+```
+TestInterestsAPI::test_with_probability_threshold
+```
 
-1. **Model Training Error Resolution:**
-   - Begin by executing `make test` to identify any issues during model training.
-   - After passing the `TestModel::test_model_build` test, use `make train` followed by `make serve-model` to train and serve the model, preparing it for upcoming steps.
+Update the API response so each interest includes a `probability` field. Verify:
+```
+make test  # test_with_probability_threshold must pass
+```
 
-2. **API Startup Issue Fix:**
-   - Resolve the Flask API startup issue with `make serve-api`.
-   - Run `make serve-api` to start a development Flask server, then open a new terminal window without shutting down the server.
+---
 
-3. **Data Return and Interests Length Correction:**
-   - Execute `make test` and resolve the `TestInterestsAPI::test_basic_response - AssertionError: Incorrect number of interests` test.
-   - Ensure the return data structure and interests array length meet the specifications, adjusting the API as necessary.
+**7. Add probability threshold filtering**
 
-4. **Response Time Optimization:**
-   - Run `make test` and fix the `TestInterestsAPI::test_basic_response - AssertionError: Request greater than 1 second` test.
-   - Enhance the API to achieve a response time of one second or less.
+Implement a feature that lets API clients filter results by a minimum probability score. For example:
 
-5. **Probability Field Inclusion:**
-   - Perform `make test` and correct the `TestInterestsAPI::test_with_probability` test.
-   - Update the API response to incorporate a 'probability' field.
+```
+GET /interests/<user_handle>?min_probability=0.5
+```
 
-6. **Probability Score Filtering Implementation:**
-   - Create a feature allowing API clients to filter results by probability score, enabling users to specify a probability threshold for more targeted results.
+Should return only interests with probability ≥ 0.5. There is no automated test for this task — demonstrate it works with a curl request.

Makefile

@@ -7,6 +7,9 @@ serve-api:
 serve-model:
 	docker compose up tf-serving
 
+ping-model:
+	curl -s http://localhost:8501/v1/models/interests-model | python3 -m json.tool
+
 train:
 	TF_CPP_MIN_LOG_LEVEL=3 python -m model.main
 

README.md

@@ -1,120 +1,92 @@
-# MLE Interview:
-
+# MLE Interview
 
 ## Setup
-1. Install [Visual Studio Code](https://code.visualstudio.com/). Also the following plugins:
-    - [Dev Containers](https://marketplace.visualstudio.com/items?itemName=ms-vscode-remote.remote-containers)
-2. Clone this repo
-3. Open the repository in VSCode (`code .`)
-4. VSCode will prompt you to install recommended extensions. Select `Yes`
-    - If you missed this step, you can install recommended extensions using the extensions sidebar (`⇧⌘X`)
-5. VSCode will prompt you to reopen the workspace in a Dev Container. Select `Yes`.
-    - If you missed this step, open the command prompt palette (`⇧⌘P`) and run the command `Remote-Containers: Rebuild and Reopen in Container`
+
+### Option A — GitHub Codespaces (recommended)
+
+Click **Code → Open with Codespaces** on the repository page. The environment builds automatically — no local installation needed.
+
+### Option B — Local Dev Container (VS Code)
+
+1. Install [Visual Studio Code](https://code.visualstudio.com/) and the [Dev Containers](https://marketplace.visualstudio.com/items?itemName=ms-vscode-remote.remote-containers) extension
+2. Install [Docker Desktop](https://www.docker.com/products/docker-desktop/)
+3. Clone this repo and open it in VS Code (`code .`)
+4. VS Code will prompt you to reopen in a Dev Container — select **Yes**
+   - If you miss the prompt: `⇧⌘P` → **Dev Containers: Rebuild and Reopen in Container**
+
+> **Note (corporate networks / Zscaler):** If your machine uses Zscaler SSL inspection, ensure your Zscaler certificate bundle exists at `~/.ssl/zscaler-bundle.pem` before building the container. The build handles this automatically.
+
+Once the container starts, dependencies are installed automatically via `make setup`. You do not need to run it manually.
 
 
 ## Make Commands
 
-Before you start, ensure you have all the necessary dependencies installed:
+### Test
 
 ```
-make setup
+make test
 ```
 
-This command will install the Python dependencies listed in `requirements.txt`. 
+Runs all tests in `./tests` with pytest. Use this to verify each task as you work through the interview.
 
-### Serve API
-
-To start the Flask API, use the following command:
+### Train
 
 ```
-make serve-api
+make train
 ```
 
-This command launches the Flask application defined in `api/app` on port 5005, accessible via `0.0.0.0`. The `--debug` flag is included to enable debug mode, which provides useful debugging information in case of errors and automatically reloads the server on code changes.
+Trains the model using `model/main.py` and exports a SavedModel to `serving/interests-model/1/`.
 
 ### Serve Model
 
-If you're using a TensorFlow model served via Docker, you can start the TensorFlow serving using:
-
 ```
 make serve-model
 ```
 
-This command uses Docker Compose to spin up a container defined in the Docker configuration, specifically targeting the TensorFlow serving service (`tf-serving`). Ensure Docker is installed and running on your system before executing this command.
+Starts TF Serving via Docker Compose on port 8501. Run this after `make train` — the model must be trained first.
 
-### Train
-
-To train the model, execute:
+### Serve API
 
 ```
-make train
+make serve-api
 ```
 
-This command runs the main training script located at `model.main`. 
+Starts the Flask development server on port 5005. Requires the model server to be running for inference endpoints to work.
 
-### Test
 
-Finally, to run the tests for your project, use:
+## API
 
 ```
-make test
+GET http://localhost:5005/interests/<user_handle>
 ```
 
-This command runs all the tests in the `./tests` directory using pytest. 
+Optional query parameters:
+- `top_k` (int) — number of interests to return (default: 10)
+- `min_probability` (float) — filter results below this probability score
 
+Example response:
 
-## API
-
-Route: GET: http://localhost:5005/interests/:userHandle
-
-API Response:
-
-```
+```json
 {
     "user_handle": "e337a675-46f5-437e-aa72-5d43643b5461",
+    "name": "Kisiza",
+    "type": "B2B",
     "interests": [
         {
             "id": "29e020bb-7a02-41db-b21f-527a8ef4dfdf",
-            "label": "Accounting technician"
+            "label": "Accounting technician",
+            "probability": 0.94
         },
         {
             "id": "012abcd1-3a6a-4803-a47e-42f46b402024",
-            "label": "Field seismologist"
+            "label": "Field seismologist",
+            "probability": 0.87
         },
         {
             "id": "a1b028dd-8464-4c63-85e8-ae29ea184fc7",
-            "label": "Designer, industrial/product"
-        },
-        {
-            "id": "686af7e4-6d58-4148-b227-3bf65ff10273",
-            "label": "Materials engineer"
-        },
-        {
-            "id": "8d1526b9-a7bf-4972-be43-7b912f149667",
-            "label": "Fashion designer"
-        },
-        {
-            "id": "d83a55a3-0143-4318-91c1-88f44ad59390",
-            "label": "Journalist, newspaper"
-        },
-        {
-            "id": "cda4441f-dba6-495c-9e2e-7429bd5e0465",
-            "label": "Therapist, music"
-        },
-        {
-            "id": "e9b23ad4-753b-4331-9cfa-525965fcf281",
-            "label": "Secretary, company"
-        },
-        {
-            "id": "ca4b03b2-0ae2-440a-afc8-5469510b19cb",
-            "label": "Commissioning editor"
-        },
-        {
-            "id": "fd3c41b8-8c15-47e2-a80d-cf3683b2d0da",
-            "label": "Copywriter, advertising"
+            "label": "Designer, industrial/product",
+            "probability": 0.81
         }
-    ],
-    "name": "Kisiza",
-    "type": "B2B"
+    ]
 }
-```
+```