- Home
- Skills
- APIs & Backend
- ad algorithm chief engineer
ad algorithm chief engineer
Autonomous driving architect playbook for sensor selection, compute budgeting, E2E migration, and safety compliance.
$10
Works with the AI Agents you already use
ad algorithm chief engineer
Example session with this skill installed
We are considering moving from a traditional modular stack to an E2E model for City NOA on Orin-X. What are the deployment risks and compute requirements?
- Read your context and instructions
- Compiled the ad algorithm chief
I will evaluate the migration to an End-to-End (E2E) architecture, focusing on compute utilization and safety guardrails.
Architecture Decision: E2E Migration for City NOA
1. Compute and Latency Budget (Estimated)
Based on a target of 30 FPS on NVIDIA Orin-X (254 INT8 TOPS):
Theoretical Demand: A complex E2E/VLA network at 1200 GFLOPs requires ~120 TOPS at 30 FPS assuming 30% utilization.
Critical Latency: The p99 latency must include exposure (33ms) + transport (10ms) + inference (60ms) + planning (20ms) + actuator (60ms) = 183ms. At 120km/h, this results in a 6.1m reaction distance.
Quantization Risk: Moving from FP16 to INT8 typically sees a 1-3% precision drop; calibration must use ODD-specific datasets to avoid long-tail failure.
2. Safety & SOTIF Evidence
| Risk Factor | Mitigation Strategy | Evidence Template |
|---|---|---|
| Black-box Uncertainty | Rule-based safety shield (RSS) post-check | references/safety-sotif.md |
| Distribution Shift | Shadow mode testing on 20k clips | references/data-loop-eval.md |
| Compute Overload | Dynamic resolution scaling / Task pruning | references/deployment-budget.md |
3. Deployment Recommendation
Conclusion: Proceed with a "Hybrid E2E" approach. Use the E2E model for trajectory proposal but maintain a hard-coded safety layer for collision avoidance (rule-based backup).
Metric Focus: Do not use mAP. Success must be measured by Mean Kilometers per Intervention (MPI) in city ODD and p99 latency jitter.
Next steps
- Run
python3 scripts/ad_budget.py tops --flops-g 1200 --fps 30 --util 0.3to refine hardware headroom. - Initialize the
templates/decision-memo.mdfor the technical review committee. - Define the auto-labeling rework rate for the new E2E training set.
Connects securely to your tools. The creator never sees your data.
About this skill
This skill provides a technical framework for autonomous driving (AD) architects and algorithm leads to evaluate perception, planning, and deployment decisions. It eliminates guesswork by enforcing specific calculation formulas, review templates, and budget accounting scripts. The tool handles the transition from traditional modular stacks to end-to-end (E2E) and Vision-Language-Action (VLA) models while maintaining functional safety and SOTIF compliance.
What it does
- Perception stack selection provides technical trade-offs between BEV, Occupancy Networks, and multi-sensor fusion strategies.
- Compute and latency budgeting calculates real-world TOPS requirements, bandwidth limits, and P99 latency chains.
- Data closed-loop planning estimates annotation costs, auto-labeling efficiencies, and regression testing thresholds.
- Safety and compliance auditing generates evidence for ASIL, SOTIF, and MRM (Minimum Risk Maneuver) reviews.
- Deployment optimization accounts for quantization effects, TensorRT constraints, and hardware utilization ratios.
How it works
- Identify the problem domain such as perception, planning, or deployment by referencing specific internal documentation modules.
- Run budget calculations using the provided scripts to generate precise hardware and data cost estimates.
- Generate formal documents using built-in templates for architecture reviews, decision memos, or budget sheets.
- Apply hard constraints to ensure every architectural recommendation includes power, latency, and data metrics.
Frameworks & tools
This skill uses Python for budget scripts and specialized templates for ISO 26262 (ASIL) and ISO/PAS 21448 (SOTIF) compliance. It targets platforms like NVIDIA Orin-X and deployment tools like TensorRT.
Why this beats prompting it yourself
Generic AI models often mistake nominal hardware TOPS for usable throughput and ignore the 30 to 70 ms latency overhead of actuators. This skill enforces rigorous math, real-world utilization ratios (20% to 40%), and specific safety templates that standard LLMs miss.
Use cases
- Select perception sensors based on FOV coverage, resolution bandwidth, and synchronization requirements.
- Migrate to end-to-end models by defining rule-based safety backups and interpretability guardrails.
- Optimize hardware costs by calculating INT8 vs FP16 trade-offs and quantization drops.
- Root cause disengagements through systematic prediction-planning-control tuning and attribution.
Known limitations
Budget calculations assume specific hardware utilization ranges and require manual input of sensor specs. Safety evidence generated is for review support and does not replace formal certification.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for APIs & Backend.
Requires manual sensor specs for accuracy. Budget scripts assume 20% to 40% utilization for complex networks. Safety outputs are evidence templates, not legal certifications.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 13 days ago