数据与 AI技能

Build persistent agents on Azure AI Foundry with the Microsoft Agent Framework Python SDK. Connect functions, search, and MCP tools, preserve conversation context, stream responses, and return data in a defined structure.

中文简介正在准备中。当前显示英语简介。

• Building an agent that preserves conversation context
• Connecting Python functions and hosted search tools
• Displaying responses as they arrive

Design agent memory that preserves facts, preferences, and conversation history across sessions. Plan storage, retrieval, and retention separately for each organization or user, with checks for unauthorized access, outdated information, conflicting records, and retrieval quality.

中文简介正在准备中。当前显示英语简介。

• Designing memory that remembers preferences across sessions
• Separating conversation history by organization and user
• Investigating retrieval of outdated or conflicting records

Provides persistent project memory for AI agents through a local MCP server, a tool interface for agents. Stores and searches architecture decisions and implementation patterns, syncs with project documentation, and includes a dashboard for viewing memory usage.

中文简介正在准备中。当前显示英语简介。

• Save architecture decisions for later reference
• Search previously recorded implementation patterns
• Retrieve design notes by key

Improve an existing AI agent by analyzing execution records, user feedback, and performance metrics. Refine prompts and workflows, compare original and revised versions through tests, and plan staged releases with monitoring and rollback when results worsen.

中文简介正在准备中。当前显示英语简介。

• Investigating recurring agent failures and user corrections
• Refining prompts, examples, and output formats
• Comparing original and revised agents with A/B tests

Route a goal through acc's scored-memory workflow, which uses evaluated stored information. Handle returned reasoning frames, submit proposals with confidence estimates, and present final answers with reference IDs before recording the eventual outcome.

中文简介正在准备中。当前显示英语简介。

• Submit a goal to acc's scored-memory workflow
• Review a returned frame and submit a proposal
• Present answers with commitment and reference IDs

Plan an A/B test to compare product changes, with a clear hypothesis, variants, sample estimates, and tracking checks. Define success metrics and stopping rules before launch, then document results and decisions.

中文简介正在准备中。当前显示英语简介。

• Planning a signup form A/B test
• Estimating sample size and test duration
• Checking event tracking before launch

Write and review SQL for reports and metrics in a data warehouse, a store of analytical data. Applies rules for explicit columns, date filters, aggregate names, and limited exploratory results, returning SQL with a brief explanation.

中文简介正在准备中。当前显示英语简介。

• Write SQL for report and dashboard metrics
• Calculate counts, sums, and rolling metrics
• Combine data from multiple tables

Find ranked crypto tokens and trader addresses by social buzz, search popularity, capital inflows, or trading profit and loss. Compare results across supported blockchains, with time windows and filters available for relevant rankings.

中文简介正在准备中。当前显示英语简介。

• Find tokens attracting social buzz
• Compare tokens by net capital inflows
• Filter trending tokens by market size

Explore World Cup match predictions, related news, and market probabilities. Adjust model factors to compare possible outcomes, then optionally hand off to Binance Agentic Wallet for prediction-market trading, where orders require an explicit choice and confirmation.

中文简介正在准备中。当前显示英语简介。

• Review AI predictions for a World Cup match
• Compare model probabilities with market probabilities
• Recalculate probabilities with adjusted factors

Decide whether an LLM needs additional training or whether retrieval and better prompts will suffice. Match available examples, preferences, or verifiable results to a training method, then assess model size and memory requirements before starting.

中文简介正在准备中。当前显示英语简介。

• Comparing fine-tuning with retrieval and prompt changes
• Choosing a training method for available data
• Assessing model size and memory requirements

Design backtests that evaluate trading strategies against historical data. Account for trading costs, avoid using future information or excluding delisted securities, and compare strategies using separate development and evaluation periods.

中文简介正在准备中。当前显示英语简介。

• Build a backtesting system for trading strategies
• Check for bias in historical performance estimates
• Compare strategies after accounting for trading costs

Prepare GRPO reinforcement learning configurations for tasks with programmatically checkable answers, such as math, code, and structured output. Design and inspect reward functions, then select training variants when runs show instability or exploit flaws in scoring.

中文简介正在准备中。当前显示英语简介。

• Preparing GRPO training for tasks with automatic correctness checks
• Designing rewards for output format and correctness
• Inspecting reward scores before training

Configure supervised fine-tuning for models that process images and text. Define which components remain unchanged, set LoRA adapters for updating selected model weights, choose image resolution limits, and investigate training runs whose evaluation results fail to improve.

中文简介正在准备中。当前显示英语简介。

• Configure training for an image-and-text task
• Decide whether to train vision layers
• Investigate training with flat evaluation scores

Design recommendation and ranking systems that select the best items for a user or situation. Organize retrieval, enrichment, filtering, scoring, selection, and background actions into six stages, then generate a runnable scaffold for your chosen stack.

中文简介正在准备中。当前显示英语简介。

• Designing a personalized content feed
• Reranking retrieved passages for RAG
• Prioritizing suggested tasks

Design machine learning workflows from data preparation through training, validation, deployment, and monitoring. Use templates and guidance to connect stages, track versions and experiments, handle failures, and plan gradual releases with rollback options.

中文简介正在准备中。当前显示英语简介。

• Designing an end-to-end machine learning pipeline
• Building reproducible training workflows
• Comparing models before production deployment

Choose a model training method and configuration from paired response preferences or individual positive and negative ratings. Covers DPO, ORPO, KTO, and SimPO, including pair construction, repeated training rounds, and troubleshooting learning settings.

中文简介正在准备中。当前显示英语简介。

• Choose a training method for paired response preferences
• Select a method for individual positive and negative ratings
• Review learning settings when DPO training loses general capability

Tune vector indexes, which organize data for similarity searches, to balance speed, memory use, and missed matches. Choose index types, adjust HNSW settings, and assess compression options using benchmarks with real queries.

中文简介正在准备中。当前显示英语简介。

• Reduce latency in similarity searches
• Tune HNSW settings to reduce missed matches
• Compare compression options for lower memory use

Prepare an evaluation framework before fine-tuning a language model. Build labeled examples, checks for each failure type, and a base-model benchmark, while calibrating model-based judges against human labels to support comparisons after training.

中文简介正在准备中。当前显示英语简介。

• Build an evaluation set from agent traces
• Create separate checks for each failure type
• Calibrate an LLM judge against human labels

Prepare datasets for language model training by formatting examples, checking conversation templates, and packing examples into training sequences. Document data sources, synthetic data ratios, and validation results in a dataset card before starting a run.

中文简介正在准备中。当前显示英语简介。

• Convert raw examples into training-ready JSONL
• Check conversation templates and assistant-only training targets
• Review example boundaries after sequence packing

Calculate portfolio risk and performance measures, including potential losses, declines from past peaks, and returns relative to risk. Use multiple indicators to support risk limits, position sizing, and ongoing monitoring.

中文简介正在准备中。当前显示英语简介。

• Measuring portfolio loss risk
• Comparing returns relative to risk
• Setting risk limits and position sizes

Turn graded execution traces and production logs into model training examples and preferred-versus-rejected response pairs. Select useful records, remove sensitive information, exclude evaluation examples, and preserve links to the original runs.

中文简介正在准备中。当前显示英语简介。

• Convert high-scoring successful runs into training examples
• Use expert-corrected failures as training examples
• Pair passing and failing responses to the same task

Export an approved fine-tuned model into a format suited to its hardware and runtime. Choose weight packaging and quantization, which reduces numerical precision, then compare outputs before and after export to check for regressions.

中文简介正在准备中。当前显示英语简介。

• Exporting a checkpoint approved for deployment
• Choosing a model format for a target GPU or laptop
• Comparing outputs before and after export

Prepare and review LoRA or QLoRA settings for adapting language models using example responses. Choose trainable layers, adapter capacity, learning rate, and memory-related options, then produce configuration values for a downstream training script.

中文简介正在准备中。当前显示英语简介。

• Preparing a LoRA configuration for supervised fine-tuning
• Reviewing rank, alpha, and learning rate settings
• Choosing LoRA or QLoRA based on GPU memory

Build searches that find similar meanings using vector databases, which store numerical representations of data. Choose search indexes, combine semantic and keyword search, and tune retrieval quality and response time for search, recommendations, or AI answer systems.

中文简介正在准备中。当前显示英语简介。

• Build search that finds documents with similar meanings
• Improve document retrieval for AI answers
• Design a recommendation engine