- Interview Q&A — AI Interviewer Questions & Answers
- These answers are self-documented after the interview — questions recalled from memory, answers researched and written post-interview for reference.
A multi-modal AI agent built for the HackerRank Orchestrate June 2026 Hackathon (24-hour challenge, June 19–20, 2026).
The agent automates insurance claim verification by analyzing both the user's written claim and photographic evidence together — using a vision-language model to produce structured, auditable decisions.
claims.csv + images
↓
Load user history & evidence requirements
↓
Encode images to base64
↓
Send to Groq Vision API (LLaMA-4-Scout)
↓
Force JSON output (10 structured fields)
↓
output.csv (44 rows, 14 columns)
- Groq API — free tier, fast inference, native vision support
meta/llama-4-scout-17b-16e-instruct— multimodal, strong JSON instruction followingtemperature=0.0— deterministic, consistent decisionsresponse_format: json_object— eliminates parsing failures- Exponential backoff — handles 429 rate limits (5 retries, 1.5x delay)
- Dynamic dataset path detection — works on any machine without path changes
git clone https://github.com/ANURAG-DASHORE/Orchestrate-Claims_Agent-HackerRank.git
cd Orchestrate-Claims_Agent-HackerRankThe dataset is not included (HackerRank's property). Clone the official repo separately:
git clone https://github.com/HackerRank/hackerrank-orchestrate-june26.gitYour folder structure should look like this:
Orchestrate-Claims_Agent-HackerRank/
├── agent.py
├── README.md
├── requirements.txt
├── .env.example
├── evaluation_report.md
└── hackerrank-orchestrate-june26/
└── dataset/
├── claims.csv
├── sample_claims.csv
├── user_history.csv
├── evidence_requirements.csv
└── images/
pip install -r requirements.txtNote: Use
pip install -r requirements.txt— NOTpip install requirements.txt
Get a free key at https://console.groq.com (takes 2 minutes).
Create a .env file in the project root:
# On Windows (cmd)
echo GROQ_API_KEY=your_actual_key_here > .env
# On Mac/Linux
echo "GROQ_API_KEY=your_actual_key_here" > .envOr rename .env.example to .env and fill in your key.
python agent.pyChoose from the menu:
=============================================
HackerRank Orchestrate Execution Menu
=============================================
1. Run Evaluation Only (20 Sample Claims)
2. Run Final Inference Only (44 Test Claims)
3. Run Both Back-to-Back
=============================================
Enter your choice (1, 2, or 3):
Each claim produces 10 fields:
| Field | Description |
|---|---|
evidence_standard_met |
true / false |
evidence_standard_met_reason |
Justification |
risk_flags |
e.g. claim_mismatch;blurry_image |
issue_type |
e.g. dent, scratch, crack |
object_part |
e.g. rear_bumper, screen |
claim_status |
supported / contradicted / not_enough_information |
claim_status_justification |
Short reasoning |
supporting_image_ids |
e.g. img_1;img_2 |
valid_image |
true / false |
severity |
low / medium / high / unknown |
The agent features a live 3-column terminal dashboard showing real-time progress broken down by claim category (Cars / Packages / Laptops):
___________________________________________________________________________________________________________________________
| |
| ____ _ _ _ |
| / __ \ | | | | | | |
| | | | |_ __ ___| |__ ___ ___| |_ _ __ __ _| |_ ___ |
| | | | | '__/ __| '_ \ / _ \/ __| __| '__/ _` | __/ _ \ |
| | |__| | | | (__| | | | __/\__ \ |_| | | (_| | || __/ |
| \____/|_| \___|_| |_|\___||___/\__|_| \__,_|\__\___| |
| |
| |
| Orchestrate - HackerRank{ orchestrate claim agent: Multi-Model Evidence Review } |
| |
| Phase: Inference (Test Claims) |
| No of Claims: 44 |
| TRUE Claims: 4 |
| FALSE Claims: 4 |
| |
| |
| | | |
| Cars: "8" | Packages: "0" | Laptops: "0" |
| TRUE Claims: "4" | TRUE Claims: "0" | TRUE Claims: "0" |
| FALSE Claims: "4" | FALSE Claims: "0" | FALSE Claims: "0" |
| | | |
| |
| [################-------------------------------------------------------------------------] 18% |
| |
| Discord: @AD1024 Github: ANURAG-DASHORE Linkedin: anurag-dashore |
---------------------------------------------------------------------------------------------------------------------------
max_retries = 5
retry_delay = 5 # seconds
for attempt in range(max_retries):
try:
response = client.chat.completions.create(...)
except Exception as e:
if "429" in str(e):
time.sleep(retry_delay)
retry_delay *= 1.5 # exponential backoff
continue- Event: HackerRank Orchestrate — June 2026 Edition
- Challenge: Multi-Modal Evidence Review
- Submitted: June 20, 2026
- Model:
meta-llama/llama-4-scout-17b-16e-instructvia Groq API - Output: 44 claims processed, results in
output.csv
| Tool | Purpose |
|---|---|
| Groq Cloud API | LLM inference — meta-llama/llama-4-scout-17b-16e-instruct vision model |
| Antigravity | Created the base architecture and initial draft of the agent |
| Google AI Studio | Prompt engineering and testing during development |
| Gemini CLI | Terminal-based AI assistance during development |
| Anthropic Claude | Re-architecture, debugging, and finalizing the complete codebase from the initial draft |
MIT