High-Throughput Office Asset Scoring Engine using Google ADK and TensorRT-LLM
Institutional investors evaluating return-to-office market trends struggle to process thousands of municipal building dossiers in real time. Build a high-throughput asset scoring engine using Google ADK and TensorRT-LLM inference runtime. Achieve sub-100ms inference latency per asset while maintaining 90% classification accuracy.
What you are building
Develop an optimized scoring pipeline with Google ADK agents powered by TensorRT-LLM serving engine to score office real estate assets.
How work is evaluated
Evaluates classification accuracy of asset scoring and verifies sub-100ms latency execution.
Datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
latency_threshold_check
Validates execution speed remains under 100ms
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
tier_classification_accuracy
Accuracy of property asset tier assignment • target: 0.9 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Learning goals
Construct a Google ADK agent pipeline utilizing Vertex AI / Gemini models for office asset grading
Deploy optimized local LLM endpoints using NVIDIA TensorRT-LLM for ultra-low latency execution
Score assets based on transit proximity, tenant credit quality, building energy rating, and flight-to-quality indicators
Benchmark end-to-end inference speed and throughput across large building portfolios
Resources
Commercial office building profile dataset containing occupancy figures, transit scores, and sustainability accreditations.
Challenge setup
Evaluates classification accuracy of asset scoring and verifies sub-100ms latency execution.
Challenge input
JSON containing occupancy_rate, transit_walk_score, lead_cert, tenant_type
Google ADK
Google Agent Development Kit for building Gemini-powered agent applications.
TensorRT-LLM
High-performance inference engine for accelerated LLM execution.
Evaluated output
JSON containing asset_tier, capital_investment_recommended, execution_time_ms
- Validates execution speed remains under 100ms
- Accuracy of property asset tier assignment • target: 0.9 • range: 0-1
- Benchmark: CRE Asset Classification Benchmark
- Tier Classification Accuracy target: 0.9
- 1 public reference case
- Python execution harness
- Python sandbox (unavailable on Versalist)
Shows the first task, up to three tools, and up to three evidence items.
View technical recipe
Configured tools
- Google ADK · Required
- Google · Optional
- TensorRT-LLM · Optional
Evaluation contract
- latency_threshold_check · Weight 1
- tier_classification_accuracy · Weight 1
Metrics
- tier_classification_accuracy · Target 0.9
Benchmarks
- CRE Asset Classification Benchmark
- general
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON containing occupancy_rate, transit_walk_score, lead_cert, tenant_type
Output: JSON containing asset_tier, capital_investment_recommended, execution_time_ms
Execution harness: Python
Evaluation tools
- Google ADK · agent_framework
- TensorRT-LLM · inference_runtime
Recipe freeze marker
No recipe freeze time is recorded.