What Optimization Examples Does Weco Offer?
Real optimization runs across agents, prompts, ML pipelines, and configs. Each one fully autonomous.
How Does Weco Optimize Code Autonomously?
Weco uses evaluation-driven code optimization. You provide source code and an evaluation script that measures performance. Weco iteratively proposes changes, tests them, and keeps the best-performing versions in a search tree.
44 examples available.
Finance & Trading (3)
Financial fraud detection pipeline optimization
End-to-end machine learning pipeline optimization to proactively identify fraudulent transactions and minimize false positives.
Metric: AUC-ROC - Improvement: +1.7% (baseline: 0.918, best: 0.9335)
Domain: Finance & Trading
Tags: Fraud, Finance, Robustness
Algorithmic pricing & bidding strategy
Makes sequential pricing decisions in competitive markets.
Metric: Score - Improvement: From zero (best: 3053156, from scratch)
Domain: Finance & Trading
Tags: Strategy, Profit
Opponent-aware market strategy optimization
Adapts decisions based on inferred competitor behavior.
Metric: Score - Improvement: +1.3% (baseline: 38671235, best: 39156328)
Domain: Finance & Trading
Tags: Strategy, Robustness
Language & Text (5)
Bias-aware toxicity detection
Detects harmful language while controlling for demographic bias and spurious correlations.
Metric: ROC AUC - Improvement: +47.3% (baseline: 0.654, best: 0.9634)
Domain: Language & Text
Tags: NLP, Robustness
Multilingual question answering (low-resource languages)
Answers questions in Hindi and Tamil with limited labeled data and brittle preprocessing.
Metric: F1 - Improvement: +21.3% (baseline: 0.505, best: 0.6128)
Domain: Language & Text
Tags: NLP, Generalization
Patent phrase similarity & deduplication
Measures semantic similarity between dense technical phrases for IP search.
Metric: Score - Improvement: Low baseline (best: 0.8501, baseline: -0.0252 — low baseline)
Domain: Language & Text
Tags: NLP, Precision
Text normalization for speech systems
Converts written text into spoken-form tokens where small errors cascade downstream.
Metric: Score - Improvement: +6.4% (baseline: 0.9335, best: 0.9931)
Domain: Language & Text
Tags: NLP, Correctness
Prompt compression for a 77-class banking intent classifier
Shrinks a 65,887-character classifier system prompt by ~95% on BANKING77 while holding accuracy at the baseline-minus-2pp threshold.
Metric: SYSTEM_PROMPT chars - Improvement: -95.1% (baseline: 65887, best: 3229)
Domain: Language & Text
Tags: NLP, Prompt, Cost
Vision, Imaging & Perception (7)
Medical X-ray abnormality detection (hospital triage)
Detects rare but critical abnormalities in chest X-rays where false negatives are costly and positives are extremely sparse.
Metric: mAP - Improvement: Low baseline (best: 0.3346, baseline: 0.0163 — low baseline)
Domain: Vision, Imaging & Perception
Tags: Vision, Safety
Skin cancer detection from dermoscopy images
Distinguishes melanoma from benign lesions under extreme class imbalance using AUROC-driven evaluation.
Metric: AUC - Improvement: +58.7% (baseline: 0.5, best: 0.7933)
Domain: Vision, Imaging & Perception
Tags: Vision, Robustness
Retail product image classification at catalog scale
Classifies products across thousands of categories with long-tail distributions and noisy merchant data.
Metric: Accuracy - Improvement: Low baseline (best: 1, baseline: 0.0089 — low baseline)
Domain: Vision, Imaging & Perception
Tags: Vision, Scalability
Histopathology cancer detection from tissue slides
Identifies cancer presence from high-resolution pathology tiles with subtle visual signals.
Metric: AUC-ROC - Improvement: +95.0% (baseline: 0.5, best: 0.9751)
Domain: Vision, Imaging & Perception
Tags: Vision, Accuracy
Chest X-ray device & line placement detection (ICU safety)
Detects catheters and tubes using multi-label prediction where misplacement can be dangerous.
Metric: AUC-ROC - Improvement: +67.5% (baseline: 0.5, best: 0.8373)
Domain: Vision, Imaging & Perception
Tags: Vision, Safety
Satellite imagery analysis for environmental monitoring
Detects subtle, rare patterns in remote-sensing imagery under changing conditions.
Metric: Dice - Improvement: +0.6% (baseline: 0.6071, best: 0.6107)
Domain: Vision, Imaging & Perception
Tags: Vision, Robustness
Document image denoising & cleanup for OCR
Restores degraded scans so downstream OCR and NLP systems don't fail.
Metric: RMSE - Improvement: -97.9% (baseline: 0.2941, best: 0.006)
Domain: Vision, Imaging & Perception
Tags: Vision, Quality
Multimodal & Scientific ML (3)
Molecular property prediction (materials science)
Predicts physical properties from molecular structure representations.
Metric: RMSLE - Improvement: -91.3% (baseline: 0.6561, best: 0.0571)
Domain: Multimodal & Scientific ML
Tags: Scientific-ML, Accuracy
RNA degradation & stability prediction
Models biological sequences where small errors change experimental outcomes.
Metric: MCRMSE - Improvement: -63.9% (baseline: 0.6421, best: 0.2321)
Domain: Multimodal & Scientific ML
Tags: Scientific-ML, Robustness
Brain tumor genetic marker prediction
Predicts molecular traits from medical images with weak labels.
Metric: Score - Improvement: +9.8% (baseline: 0.5, best: 0.5492)
Domain: Multimodal & Scientific ML
Tags: Multimodal, Accuracy
Inference, Forecasting & Estimation (3)
Road network inference from travel queries
Infers unknown road distances using shortest-path query feedback.
Metric: Score - Improvement: From zero (best: 47350549621, from scratch)
Domain: Inference, Forecasting & Estimation
Tags: Inference, Accuracy
Demand & system state forecasting
Estimates hidden system states from noisy observations over time.
Metric: Score - Improvement: +28,613% (baseline: 125937215, best: 36160033559)
Domain: Inference, Forecasting & Estimation
Tags: Inference, Robustness
Industrial parameter estimation
Infers latent parameters governing system behavior.
Metric: Score - Improvement: From zero (best: 238902068, from scratch)
Domain: Inference, Forecasting & Estimation
Tags: Inference, Accuracy
Forecasting & Business ML (4)
NYC taxi fare prediction (pricing engine)
Predicts fares from noisy trip records while avoiding leakage and unstable features.
Metric: RMSE - Improvement: -6.8% (baseline: 24.7876, best: 23.1066)
Domain: Forecasting & Business ML
Tags: Forecasting, Accuracy
Retail demand forecasting & recommendations
Predicts future purchases from massive transactional histories with delayed rewards.
Metric: MAP - Improvement: From zero (best: 0.012, from scratch)
Domain: Forecasting & Business ML
Tags: Forecasting, Revenue
Customer segmentation from weak signals
Segments customers using noisy, incomplete attributes with limited labels.
Metric: Silhouette - Improvement: +145.3% (baseline: 0.165, best: 0.4047)
Domain: Forecasting & Business ML
Tags: Forecasting, Robustness
Synthetic tabular benchmarking (pipeline stress tests)
Stress-tests feature engineering, validation, and leakage handling on controlled data.
Metric: R² - Improvement: Low baseline (best: 0.8169, baseline: -0.0632 — low baseline)
Domain: Forecasting & Business ML
Tags: Forecasting, Validation
Time-Series, Control & Signals (4)
ICU ventilator pressure control modeling
Predicts control signals from noisy physiological time-series where stability matters more than raw accuracy.
Metric: MAE - Improvement: -98.8% (baseline: 17.5766, best: 0.2173)
Domain: Time-Series, Control & Signals
Tags: Time-Series, Stability
EEG-based harmful brain activity detection
Detects rare clinical events from high-frequency, multichannel EEG data.
Metric: Score - Improvement: -43.4% (baseline: 1.3915, best: 0.7879)
Domain: Time-Series, Control & Signals
Tags: Time-Series, Safety
Pulmonary function decline forecasting
Forecasts disease progression from sparse, longitudinal patient records.
Metric: Score - Improvement: +70.6% (baseline: -24.7981, best: -7.2826)
Domain: Time-Series, Control & Signals
Tags: Time-Series, Robustness
Environmental sound classification
Tags background audio for moderation and search under noisy conditions.
Metric: Score - Improvement: Low baseline (best: 0.7021, baseline: 0.0166 — low baseline)
Domain: Time-Series, Control & Signals
Tags: Audio, Accuracy
Logistics, Delivery & Transportation (2)
City transportation authority route planning
Optimizes routes across a large urban road network under congestion and structural constraints.
Metric: Score - Improvement: From zero (best: 34017823, from scratch)
Domain: Logistics, Delivery & Transportation
Tags: Logistics, Efficiency
Autonomous driving decision optimization (Toyota challenge)
Optimizes decisions in a realistic automotive control environment.
Metric: Score - Improvement: From zero (best: 48561411301, from scratch)
Domain: Logistics, Delivery & Transportation
Tags: Planning, Safety
Manufacturing, Factories & Operations (3)
Factory job-shop scheduling
Assigns jobs to machines to minimize idle time, delays, and bottlenecks.
Metric: Score - Improvement: From zero (best: 49360632, from scratch)
Domain: Manufacturing, Factories & Operations
Tags: Scheduling, Throughput
Semiconductor layout & assembly optimization
Optimizes placement and sequencing of tightly constrained components to reduce interference and maximize manufacturability.
Metric: Score - Improvement: +37.6% (baseline: 30478055, best: 41943886)
Domain: Manufacturing, Factories & Operations
Tags: Manufacturing, Efficiency
Warehouse robot fleet coordination
Coordinates many robots to avoid conflicts and complete tasks efficiently.
Metric: Score - Improvement: From zero (best: 9014000, from scratch)
Domain: Manufacturing, Factories & Operations
Tags: Planning, Coordination
Infrastructure, Utilities & Networks (4)
Power grid & utility network design
Designs network topology to balance cost, redundancy, and reliability.
Metric: Score - Improvement: From zero (best: 117403440265, from scratch)
Domain: Infrastructure, Utilities & Networks
Tags: Network, Cost
Telecom & data-center network topology planning
Optimizes connectivity while meeting budget and performance constraints.
Metric: Score - Improvement: +2.5% (baseline: 4619973526, best: 4734320999)
Domain: Infrastructure, Utilities & Networks
Tags: Network, Efficiency
Cell-tower / sensor coverage planning
Places minimal infrastructure to fully cover a geographic region.
Metric: Score - Improvement: +16.1% (baseline: 69301508, best: 80459999)
Domain: Infrastructure, Utilities & Networks
Tags: Covering, Cost
Energy grid load & resource planning
Allocates limited resources across future time periods to avoid peaks.
Metric: Score - Improvement: +2.7% (baseline: 45500, best: 46734)
Domain: Infrastructure, Utilities & Networks
Tags: Planning, Efficiency
Video, Media & Streaming (1)
AV1 video encoder param search (SVT-AV1)
Searches the SVT-AV1 reference encoder’s preset, CRF, tune, film-grain, and svtav1-params space to find a Pareto-better encode of a 1080p Tears of Steel clip at VMAF ≥ 95. Beats a hand-tuned 4-config Pareto sweep — automated search recovers the streaming-engineer optimum in 12 steps and pushes past it via non-obvious svtav1-params (scd=0, tune=1 PSNR-aligned, max lookahead, single-tile).
Metric: File size (KB) at VMAF≥95, ≤60s budget - Improvement: -10.9% (baseline: 789.79, best: 703.77)
Domain: Video, Media & Streaming
Tags: Video, Codec, Compression
Kernel Optimization (5)
Causal self-attention optimization (GPT-style LLMs)
Generates a faster GPU kernel for causal self-attention, reducing per-token inference latency in GPT-style language models.
Metric: Speedup - Improvement: +47.7% (baseline: 0.9992, best: 1.4759)
Domain: Kernel Optimization
Tags: Latency, LLMs, Kernels
Transformer block optimization (MiniGPT-style)
Optimizes a full Transformer block by improving how attention, normalization, and feedforward ops execute together.
Metric: Speedup - Improvement: +34.1% (baseline: 1.0021, best: 1.3436)
Domain: Kernel Optimization
Tags: Latency, LLMs, Kernels
GELU activation optimization (Transformer workloads)
Produces a faster GPU implementation of GELU, a core activation used throughout Transformer-based models.
Metric: Speedup - Improvement: +283.1% (baseline: 1.01, best: 3.8696)
Domain: Kernel Optimization
Tags: Throughput, LLMs, Kernels
3D transposed convolution optimization (video models)
Optimizes a heavy 3D convolution used in video generation and volumetric vision pipelines.
Metric: Speedup - Improvement: +175.1% (baseline: 0.9989, best: 2.7477)
Domain: Kernel Optimization
Tags: Latency, Video, Kernels
State-space model optimization (Mamba-style architectures)
Speeds up sequence-processing kernels used in state-space models such as Mamba, a Transformer alternative for long-context workloads.
Metric: Speedup - Improvement: +24.6% (baseline: 0.983, best: 1.2249)
Domain: Kernel Optimization
Tags: Throughput, LLMs, Kernels