Data and AI
33 use cases, 11 specialists
AI engineer
Sales forecasting with AI and machine learning
Writes and runs Python code on a sales export from Excel or 1C, trains a forecasting model, and saves the forecast with a chart and an accuracy estimate.
Classifying reviews by sentiment: AI agent writes the code
Writes code that labels reviews and customer requests from a table by sentiment and topic and summarizes the results by category in Excel.
AI agent for building a store recommendation system
Writes code for "frequently bought together" recommendations from order history, checks their quality on held-out data, and delivers a module with a description.
Data engineer
AI agent for building ETL pipelines
Writes code that loads data from 1C, CRM, databases, and files into a data warehouse so that a rerun does not create duplicates and a change in the source structure triggers an alert.
Data quality checks with AI
Writes and runs table checks for missing values, duplicates, freshness, and discrepancies with the source and produces a report for each table.
Designing a data warehouse and data marts: AI assistant
Proposes warehouse layers based on a description of the sources and the reports needed and writes dbt models and data mart SQL for reporting systems.
Data visualization engineer
AI for creating charts and graphs
Picks the chart type that fits the question for an Excel or CSV table, builds it with code, and saves it as an image or a file with labeled axes.
Interactive charts in D3 and Vega with AI
Writes an interactive chart with tooltips, filters, and keyboard control as an HTML page that opens in a browser.
AI chart review: honest axes and clear colors
Looks at a finished chart or report, finds truncated axes, dual scales, indistinguishable colors, and a poorly chosen chart type, and proposes a corrected version.
Data remediation engineer
Finding errors in large tables: AI groups similar ones
Reduces thousands of broken rows to a few groups of similar errors and shows examples and the likely cause for each group.
AI for normalizing addresses, phone numbers, and names
Writes a correction rule for each group of errors, brings addresses, phone numbers, and counterparty names into a uniform format, and produces a corrected file with a report on the changes.
Data correction with a before-and-after log: AI agent
Prepares corrections in a staging table, keeps a log of old and new values for each row, sets aside rows that cannot be fixed for manual review, and verifies that no record has been lost.
Email processing engineer
AI for analyzing business correspondence
Reconstructs email threads from a mail export, removes repeated quotes and signatures, and writes a summary: who proposed what and what was agreed.
Tasks and agreements from email into a spreadsheet with AI
Finds promises, deadlines, and assignments in emails, ties them to specific people, and compiles them into an Excel table with a link to the original email.
AI agent for parsing incoming emails into a CRM
Parses incoming emails and attachments, extracts company details, amounts, and contacts, and saves them to JSON or a table for loading into a CRM.
Knowledge graph engineer
Knowledge graph from documents: AI agent links the facts
Extracts entities and the relationships between them from documents, builds a graph in Neo4j, and stores a link to the source file with every fact.
AI for finding contradictions in documents
Compares statements from different sources, flags those that contradict each other, and shows both versions with links, without overwriting either one.
What a change will affect: relationship analysis with AI
Uses the graph to find everything that depends on the changed document or service and produces a list of affected nodes with their relationship chains.
AI document search engineer (RAG)
Chatbot for a company knowledge base: AI search across documents
Splits company documents into chunks, builds hybrid search by meaning and by keyword, and puts together code that answers questions with a reference to the source.
Preparing documents for AI search
Chooses a chunking method to suit the document type (by headings, by meaning, or by size) and tags the chunks with metadata for precise filtering.
RAG system quality evaluation: AI engineer measures accuracy
Assembles a reference set of questions and measures retrieval accuracy and answer correctness before and after every change to the settings.
Search quality engineer
AI assistant for tuning site and catalog search
Configures analyzers, synonyms, typo tolerance, and field weights in Elasticsearch or OpenSearch so that products are found the way shoppers type them.
Search does not find products: AI analyzes the query log
Analyzes the search query log, singles out queries with empty results and reformulations, and suggests synonyms and index changes.
Evaluating search relevance with nDCG using AI
Assembles a reference set with ratings from the query log, calculates nDCG and MRR, and compares search settings before and after a change.
Language model fine-tuning engineer
Fine-tuning a language model on your own data: AI assistant
Prepares and checks a set of examples for fine-tuning, locks in the data, tokenizer, and settings, and writes the code for a trial run.
AI for analyzing model training failures
Writes a failure report from the run's logs and metrics: observations, the likely cause, and the next minimal experiment with a stopping condition.
Is the new model version ready: AI check before release
Compares the new version with the original on identical tests, verifies the integrity of the saved weights by checksums, and writes up a release record.
Machine learning model auditor
Scoring model validation: AI assistant for auditors
Reproduces the model's sample and training from the documentation, compares the results with the original ones, and produces a report on the findings with an assessment of their importance.
Monitoring data drift and model stability with AI
Calculates the PSI stability index by feature, plus quality and calibration metrics by period and segment, and flags where the model has started to degrade.
Machine learning model explainability: SHAP analysis with AI
Builds SHAP and partial dependence plots and explains in plain language which features affect the model's decisions and how, including for different groups of people.