ghaymah-hiring/ghaymah-genai-exam
1
0
نسخ 1
الشفرة المشكلات طلبات السحب Actions Packages المشاريع الإصدارات الويكي النشاط

Finalize GenAI exam: simplify qabilah+q2, add q5 LangChain agent

تصفح المصدر
هذا الالتزام موجود في:
Ghaymah
2026-08-31 06:52:52 +00:00
الأصل 08cb1ef544
التزام 52ac62ab40
7 ملفات معدلة مع 172 إضافات و100 حذوفات
تنزيل ملف التصحيح تنزيل ملف الاختلاف توسيع جميع الملفات طي جميع الملفات

47
README.md
عرض الملف

@@ -15,8 +15,8 @@ You are interviewing to join the team that builds **agents like Cumin** — the
1. Introduce yourself to the Arabic tech ecosystem (**Qabilah** + **Mortakaz**).
2. Design a **memory module** for an agent like Cumin, and show how to use it.
3. Plan how to **fine-tune an Arabic TTS or STT** model.
4. Sign up to **Ghaymah** and **deploy a simple static website**.
3. Propose an **Arabic TTS or STT** model + dataset.
4. Sign up to **Ghaymah**, **deploy a static website**, and build a **LangChain agent** on Ghaymah's GenAI services.
## What we are evaluating
@@ -24,8 +24,9 @@ You are interviewing to join the team that builds **agents like Cumin** — the
|---|---|
| Arabic tech ecosystem awareness | Qabilah + Mortakaz tasks |
| Agent engineering intuition (memory, retrieval, storage trade-offs) | Q1 |
| Arabic NLP depth (data, diacritization, evaluation, fine-tuning) | Q2 |
| Shipping — can you actually deploy something live | Q3 + Q4 |
| Arabic NLP depth (model + dataset selection) | Q2 |
| Shipping — deploy something live | Q3 + Q4 |
| Agent building with real tooling (LangChain) | Q5 |
## The stack you will touch
@@ -34,19 +35,20 @@ You are interviewing to join the team that builds **agents like Cumin** — the
| **Qabilah (قبيلة)** | Arabic tech social / community platform | https://qabilah.com |
| **Mortakaz (مُرتكز)** | Directory of ~686 Arab-built products | https://mortakaz.com |
| **Ghaymah (غيمة)** | Arabic cloud platform (PaaS) | https://ghaymah.systems |
| **Ghaymah Dashboard** | Where you sign up and deploy | https://deploy.ghaymah.systems |
| **Ghaymah Dashboard** | Where you sign up, get AI models, and deploy | https://deploy.ghaymah.systems |
## Time budget (60 min total)
| Task | Folder | Budget |
|---|---|---|
| Read this README + set up | — | 5 min |
| Qabilah profile + ecosystem note | `common-qabilah/` | 8 min |
| Qabilah profile + check-in | `common-qabilah/` | 3 min |
| Mortakaz discovery + integration proposal | `common-mortakaz/` | 10 min |
| Q1 — Cumin memory module design | `q1-cumin-memory/` | 15 min |
| Q2 — Arabic TTS or STT fine-tune plan | `q2-arabic-tts-stt/` | 15 min |
| Q3 — Ghaymah signup | `q3-ghaymah-signup/` | 5 min |
| Q4 — Deploy a static site on Ghaymah | `q4-static-site-deploy/` | 12 min |
| Q2 — Arabic TTS/STT model + dataset proposal | `q2-arabic-tts-stt/` | 6 min |
| Q3 — Ghaymah signup | `q3-ghaymah-signup/` | 4 min |
| Q4 — Deploy a static site on Ghaymah | `q4-static-site-deploy/` | 8 min |
| Q5 — LangChain agent on Ghaymah's GenAI services | `q5-llm-agent/` | 9 min |
| **Total** | | **≈ 60 min** |
> ⏱️ Be honest about your time. A focused, complete 60-minute submission beats a half-finished 2-hour one. We look at commit timestamps.
@@ -56,7 +58,7 @@ You are interviewing to join the team that builds **agents like Cumin** — the
1. Work **inside this repo**. Each question has its own folder with a task description and an answer template. Fill in your answers **in place**.
2. Do not rename or move the question folders.
3. Never commit real credentials, tokens, or passwords — use placeholders like `<REDACTED>`.
4. For Q4, commit your actual site source under `q4-static-site-deploy/site/`.
4. For Q4, commit your actual site source under `q4-static-site-deploy/site/`; for Q5 under `q5-llm-agent/`.
5. You may use any public resources (docs, model cards, papers, blogs) — cite them inline.
## Repository structure
@@ -71,14 +73,18 @@ ghaymah-genai-exam/
├── q1-cumin-memory/
│ └── cumin-memory-module.md ← task + answer template
├── q2-arabic-tts-stt/
│ └── arabic-tts-stt-finetune.md ← task + answer template
│ └── arabic-tts-stt-proposal.md ← task + answer template
├── q3-ghaymah-signup/
│ └── ghaymah-signup.md ← task + answer template
└── q4-static-site-deploy/
├── static-site-deploy.md ← task + answer template
└── site/ ← put your site files here
├── index.html
└── style.css
├── q4-static-site-deploy/
│ ├── static-site-deploy.md ← task + answer template
│ └── site/
│ ├── index.html
│ └── style.css
└── q5-llm-agent/
├── llm-agent.md ← task + answer template
├── agent.py ← starter LangChain agent
└── requirements.txt
```
## How to submit (GitPasha)
@@ -89,18 +95,19 @@ ghaymah-genai-exam/
git add . && git commit -m "q1: memory module design"
```
3. `git push origin main` when done.
4. Verify on GitPasha that every file (including your Q4 site source) is present.
4. Verify on GitPasha that every file (including your Q4 site source and Q5 code) is present.
## Scoring rubric
| Question | Weight | What “excellent” looks like |
|---|---|---|
| common-qabilah | 10% | Real profile URL + a note showing genuine exploration, not copy-paste |
| common-qabilah | 5% | Real profile URL |
| common-mortakaz | 15% | Two concrete products, technically-grounded integration with Ghaymah's AI stack, honest challenges |
| Q1 memory module | 30% | Clear memory taxonomy, concrete storage choices, sensible write/read policy, runnable pseudocode, Ghaymah integration |
| Q2 TTS/STT | 25% | Correct Arabic data/preprocessing/eval, realistic training config, deployable plan |
| Q1 memory module | 25% | Clear memory taxonomy, concrete storage choices, sensible write/read policy, runnable pseudocode, Ghaymah integration |
| Q2 TTS/STT | 15% | A well-justified model + dataset with correct Arabic specifics (dialect, diacritization, license) |
| Q3 signup | 5% | Confirmed signup + thoughtful platform observations |
| Q4 static site | 15% | A live URL that resolves + source committed + a short write-up |
| Q5 LangChain agent | 20% | A working agent wired to Ghaymah's LLM endpoint, with at least one tool and clear code |
---

30
common-qabilah/qabilah-ecosystem.md
عرض الملف

@@ -1,20 +1,14 @@
# Qabilah (قبيلة) — Ecosystem Introduction
# Qabilah (قبيلة) — Profile & Ecosystem Check-in
⏱ **Budget: ~8 minutes** · Weight: 10%
⏱ **Budget: ~3 minutes** · Weight: 5%
## The task
[Qabilah](https://qabilah.com) is an Arabic tech social & community platform — think of it as the Arabic-speaking tech community's home. Before you design agents for this ecosystem, you should know who lives in it.
[Qabilah](https://qabilah.com) is an Arabic tech social & community platform — think of it as the Arabic-speaking tech community's home. Before you build agents for this ecosystem, you should know who lives in it.
1. Go to https://qabilah.com and **create a profile** (if you don't already have one). Use a real name and a one-line bio about what you build.
2. Browse the feed / communities. Find **three Arabic AI or GenAI projects, people, or companies** that interest you.
3. Write a short note (150–250 words) answering: **what is Qabilah, and what role does it play in the Arabic tech ecosystem?**
## What to submit
Fill in the template below, in this file. Put your profile URL in section 1 and your note in section 2.
---
2. Record your profile URL below.
3. (Optional, one line) In your own words: what is Qabilah?
## ANSWER — Qabilah
@@ -24,16 +18,6 @@ Fill in the template below, in this file. Put your profile URL in section 1 and
https://qabilah.com/profile/<your-handle>/posts
```
### 2. Ecosystem note (150–250 words)
### 2. One line: what is Qabilah?
> _Write here. What is Qabilah? What problem does it solve for Arabic-speaking builders? How does it connect people/projects/funding? You may write in Arabic or English._
<!-- replace with your note -->
### 3. Three Arabic AI projects / people I found
| # | Name | What it does | Why it interested me |
|---|---|---|---|
| 1 | | | |
| 2 | | | |
| 3 | | | |
> _optional — one sentence_

57
q2-arabic-tts-stt/arabic-tts-stt-finetune.md
عرض الملف

@@ -1,57 +0,0 @@
# Q2 — Fine-tuning an Arabic TTS or STT
⏱ **Budget: ~15 minutes** · Weight: 25%
## Choose ONE option
- **Option A — TTS.** Fine-tune an open Arabic (or multilingual) text-to-speech model — e.g. Coqui **XTTS-v2**, **VITS**, or **StyleTTS2** — on a single-speaker Arabic dataset.
- **Option B — STT.** Fine-tune **Whisper** (e.g. `whisper-small`, `large-v3`) on Arabic speech — e.g. Mozilla **Common Voice `ar`**, **MGB-2**, or a custom corpus.
## Produce a runnable guide covering
1. **Model & dataset choice** — justify it; cite dataset sizes, license, and dialect (MSA vs. Egyptian vs. Gulf vs. Maghrebi).
2. **Data preprocessing** — audio resampling/normalization, text normalization, and the Arabic-specific gotchas: **diacritization (تشكيل)**, letter variants (أ/إ/آ, ة/ه, ى/ي), and (for STT) whether to strip diacritics before computing WER. Train/val split.
3. **Training setup** — framework (e.g. Hugging Face Trainer, Coqui TTS, NeMo), **LoRA vs. full fine-tune**, key hyperparameters (batch size, LR, epochs, warmup), hardware, and a rough time/cost estimate.
4. **Evaluation** — for STT: **WER/CER** (and how to compute them fairly for Arabic); for TTS: **MOS**, **speaker similarity (SECS)**, and using an ASR to measure intelligibility. Include the actual metric definitions/commands.
5. **Deployment on Ghaymah** — GPU sizing, container vs. function, serving stack, latency, and how the model is served behind an API.
6. **Risks & mitigations** — dialect coverage, hallucination, diacritic fidelity, MSA vs. dialect mismatch, data licensing.
Include **real code snippets** (Python / Hugging Face / CLI). Cite sources inline.
## Deliverable
Fill in the template below. State clearly which option you chose.
---
## ANSWER — Arabic TTS/STT Fine-tune
**Option chosen:** A / B (delete one)
### 1. Model & dataset choice
> _write here_
### 2. Data preprocessing
> _write here + code snippets_
### 3. Training setup
> _write here + code/config snippets_
### 4. Evaluation
> _write here + metric definitions/commands_
### 5. Deployment on Ghaymah
> _write here_
### 6. Risks & mitigations
| Risk | Mitigation |
|---|---|
| | |
| | |
| | |

39
q2-arabic-tts-stt/arabic-tts-stt-proposal.md Normal file
عرض الملف

@@ -0,0 +1,39 @@
# Q2 — Propose an Arabic TTS or STT Model + Dataset
⏱ **Budget: ~6 minutes** · Weight: 15%
## Choose ONE option
- **Option A — TTS.** Propose a text-to-speech model to fine-tune for Arabic.
- **Option B — STT.** Propose a speech-to-text model to fine-tune for Arabic.
## The task
Propose **one model** and **one dataset**, with a short justification. No training, no running code — just a well-reasoned pick. (Short code/CLI snippets are allowed but not required.)
Cover:
1. **Model** — name it, and say why: architecture, license, multilingual/Arabic support.
2. **Dataset** — name it, and say why: size, dialect (MSA vs. Egyptian vs. Gulf vs. Maghrebi), license, where to get it.
3. **Why this pairing works for Arabic** — 1–2 sentences (e.g. diacritization handling, dialect coverage).
4. **One key risk** — 1 sentence.
## ANSWER — Arabic TTS/STT Proposal
**Option chosen:** A / B (delete one)
### Proposed model
> _write here_
### Proposed dataset
> _write here_
### Why this pairing works for Arabic
> _write here_
### One key risk
> _write here_

47
q5-llm-agent/agent.py Normal file
عرض الملف

@@ -0,0 +1,47 @@
"""
Q5 starter — a minimal LangChain agent on Ghaymah's GenAI services.
1. Get base_url + API key + model ID from https://deploy.ghaymah.systems
2. pip install -r requirements.txt
3. python agent.py "what is 12 * 7 plus 4?"
"""
import os
import sys
from langchain_openai import ChatOpenAI
from langchain.agents import AgentExecutor, create_tool_calling_agent
from langchain.tools import tool
from langchain_core.prompts import ChatPromptTemplate
# --- Ghaymah GenAI config (fill these in / use env vars) ---
BASE_URL = os.getenv("GHAYMAH_BASE_URL", "https://<ghaymah-llm-endpoint>/v1")
API_KEY = os.getenv("GHAYMAH_API_KEY", "<redacted>")
MODEL = os.getenv("GHAYMAH_MODEL", "<model-id-from-dashboard>")
llm = ChatOpenAI(base_url=BASE_URL, api_key=API_KEY, model=MODEL, temperature=0)
@tool
def calculator(expression: str) -> str:
"""Evaluate a simple arithmetic expression and return the result."""
try:
return str(eval(expression, {"__builtins__": {}}, {}))
except Exception as e: # noqa: BLE001
return f"error: {e}"
tools = [calculator]
prompt = ChatPromptTemplate.from_messages(
[
("system", "You are a helpful Arabic/English assistant. Use the calculator tool when you need arithmetic."),
("human", "{input}"),
("placeholder", "{agent_scratchpad}"),
]
)
agent = create_tool_calling_agent(llm, tools, prompt)
executor = AgentExecutor(agent=agent, tools=tools, verbose=True)
if __name__ == "__main__":
q = sys.argv[1] if len(sys.argv) > 1 else "ما حاصل 12 × 7 + 4؟"
print(executor.invoke({"input": q})["output"])

50
q5-llm-agent/llm-agent.md Normal file
عرض الملف

@@ -0,0 +1,50 @@
# Q5 — Build a simple LLM agent with LangChain (on Ghaymah's GenAI services)
⏱ **Budget: ~9 minutes** · Weight: 20%
## The task
Build a small **LLM agent** using **LangChain**, and point it at **Ghaymah's GenAI services** (the ready-to-use LLM models you saw in the dashboard).
Requirements:
1. Use LangChain (Python) with an **OpenAI-compatible** client pointed at Ghaymah's inference endpoint. Get the **base URL**, **API key**, and a **model ID** from https://deploy.ghaymah.systems (see the AI / models section).
2. Give the agent **at least one tool** — e.g. a calculator, a "save note" memory tool, or a tiny retrieval tool.
3. (Optional, +bonus) Wire it to the static site from **Q4** — the page calls your agent's endpoint and shows the reply.
There's a starter `agent.py` + `requirements.txt` in this folder you may extend or replace.
## Deliverable
Fill in the template below and commit your code in this folder.
---
## ANSWER — LangChain agent
### 1. What my agent does
> _one or two lines_
### 2. Architecture (LangChain components)
| Component | What I used |
|---|---|
| Model (LLM) | |
| Tool(s) | |
| Prompt | |
| Executor | |
### 3. How it uses Ghaymah's GenAI services
> _base URL pattern, model ID, how you got the key (do not paste the key)_
### 4. Code
> _commit your code in this folder (agent.py etc.) and reference it here_
### 5. How to run / deploy it
```bash
# commands
```

2
q5-llm-agent/requirements.txt Normal file
عرض الملف

@@ -0,0 +1,2 @@
langchain
langchain-openai