Below are eight scenarios that you can test on your own files, with ready-made queries and limitations that are important to remember.
<a></a>1. Turn notes into a working document
How it is now. After a discussion, there are messages in a messenger, bullet points in a notebook, and several voice remarks. To get an instruction or a brief from them, you first have to reconstruct the logic of the conversation yourself.
What to ask AI. "Below are my notes after the meeting. Compile an instruction for a new employee from them: goal, sequence of actions, responsible parties, deadlines, and unresolved issues. Do not add facts from yourself, mark omissions as 'needs clarification'."
What result. A structured draft instead of scattered notes. It can be proofread, supplemented with missing data, and passed on to a colleague.
Pitfalls. If there is no deadline or responsible party in the notes, the neural network may try to fill the gap with a plausible wording. Therefore, it is better to include a prohibition on guessing directly in the query.
<a></a>2. Conduct a search and compile conclusions with links
How it is now. Research begins with a dozen tabs. One contains documentation, another a forum discussion, and a third an old article that no longer corresponds to the current version of the product.
What to ask AI. "Compare the current conditions of [five Russian video conferencing services for a team of 30 people - replace for your research]. Use official websites and documentation, indicate the date of each source, compile a table by price, recording restrictions, and payment methods. If the data differs, show it separately."
What result. A brief note, a comparative table, and links to sources. No need to manually transfer facts from each tab to a separate document.
Pitfalls. For important comparisons, you can immediately ask Iskra to conduct an additional check: use official sources, verify dates, and separately note discrepancies between tariffs or product versions.
<a></a>3. Compare several PDFs or document versions
How it is now. Two editions of a regulation, commercial offer, or technical description are opened side-by-side, and differences are sought visually. A rearranged paragraph can easily be mistaken for a new one, and a small change in a number can be missed.
What to ask AI. "Compare these two PDFs. Create a table of changes: section, old wording, new wording, possible consequence. Do not consider text rearrangement without a change in meaning as a substantive edit. For each item, indicate the page in both files."
What result. A map of changes that shows what has actually changed and which places need to be read carefully.
Pitfalls. A scan without a text layer is recognized worse than a regular PDF. If the PDF consists of scans, you can immediately ask Iskra to additionally recheck numbers, formulas, and small text, and separately mark uncertainly recognized fragments in the result.
<a></a>4. Transfer data from a scan or screenshot to a table
How it is now. A price list, invoice, or list of participants arrives as a photo. Data has to be retyped, and then errors in amounts, phone numbers, and names have to be found.
What to ask AI. "Extract rows from these images into a table with columns 'name', 'quantity', 'price', and 'amount'. Do not guess illegible values: put a 'check' mark. At the end, verify the arithmetic and list the rows where the total does not match."
What result. A table that can be downloaded and further processed, plus a separate list of questionable places for manual verification.
Pitfalls. Glare, camera tilt, and handwritten notes reduce accuracy. The more important the document, the more dangerous it is to accept a neatly formatted result as proof of its correctness.
<a></a>5. Clean and merge tables
How it is now. Two exports about the same thing look different: in one, the phone number is recorded starting with an eight, in the other with '+7', column names do not match, and some rows are repeated. Formulas and manual filters have to be reassembled for each task.
What to ask AI. "Merge two exports of event participants. Normalize phone numbers and emails, find duplicates, but do not delete anything automatically. Return the final table, a separate sheet with suspected duplicates, and briefly describe the rules by which you found them."
What result. A ready-made table and a clear transformation log. For a file with several hundred rows, initial processing can be obtained in minutes instead of manually checking each row.
Pitfalls. A matching name does not necessarily mean it's the same person, and different addresses do not guarantee that the people are different. Ambiguous merges must be confirmed by the user.
<a></a>6. Turn a meeting recording into a protocol
How it is now. The meeting was recorded, but no one wants to re-listen to an hour of conversation for three decisions. As a result, agreements remain in the participants' memory, and a week later, everyone remembers them differently.
What to ask AI. Attach audio or video and write: "Transcribe, then prepare a protocol: decisions, tasks, responsible parties, deadlines, and open questions. For each item, indicate the timestamp. Mark names and special terms you are unsure about for verification."
What result. The meeting text and a short protocol, which can be used to immediately assign tasks. The long recording does not need to be re-listened to in its entirety.
Pitfalls. Names, abbreviations, and the speech of several people may be inaccurately recognized.
<a></a>7. Assemble a presentation from source materials
How it is now. The content of the report is already ready, but it still needs to be laid out on slides, formulations shortened, headings invented, and a suitable illustration found.
What to ask AI. "Based on this report, prepare an eight-slide presentation for the manager. On the first slide, formulate the main conclusion, then show three facts, the problem, solution options, and the next step. Take all numbers only from the file. For the cover, create a calm horizontal illustration without text."
What result. A draft presentation with ready-made logic and visual material. The human only needs to check the facts, shorten overloaded slides, and bring the design to the company's template.
Pitfalls. The generated image will be raster, not a Figma layout or vector corporate style. Images are created through an external provider and consume the service's overall limit.
<a></a>8. Ask about your own documents, rather than starting over
How it is now. In each new dialogue, you have to re-attach the regulations, product description, and examples of good answers. Without context, the neural network responds with general words and does not know internal rules.
What to ask AI. Instructions, regulations, presentations, and other working documents can be uploaded to IskraBot's memory. After that, the query looks simpler: "According to our event preparation rules, create a webinar launch checklist. For each item, show which document the requirement was taken from. If the answer is not in the documents, state so."
What result. AI responds in the context of the selected materials and shows sources. This mode is useful not for a single question, but for repetitive work with a single set of documents.
Pitfalls. The neural network will not determine itself that the regulation is outdated. Memory needs to be updated along with the source documents, otherwise the answer will be accurate but based on old rules.
<a></a>What is important when choosing an AI service for work in Russia
The quality of the answer depends on the AI model, but in daily work, there are more mundane limitations.
Access without workarounds. IskraBot does not require a VPN. The service opens in a regular browser, so you can start without separate connection instructions.
Payment in rubles. The personal tariff is paid with a Russian card. For organizations, an invoice, contract, and closing documents are available through EDO.
Working with files. The task does not end with a message in the chat. The platform reads common documents and tables, and the agent can prepare a new file that the user downloads and continues to edit.
Access without VPN and payment in rubles are also offered by other Russian services. IskraBot's focus is on a completed file task: accept source files, return an editable file, save context in memory, and connect a skill or integration if necessary.
<a></a>From personal task to team work
For yourself. In IskraBot, you can start with chat, search, and file operations. The free Start tariff is sufficient to test a basic scenario on your own PDF or table.
For the team. AI for business adds unified access, a common token pool, memory for regulations, and pluggable integrations. Cloud data of organizations is stored in Russia, and upon request, the processing of organization requests can be directed to Russian servers.
For a large company. The platform can be deployed in your own environment. What external challenges will remain depends on the tested configuration of a specific installation; masking and auditing are connected separately.
<a></a>How to start
To test IskraBot in your work, take one small task - comparing two documents, cleaning an export, or analyzing a meeting recording - and run it on your own files. You can start with the free Start tariff at iskrabot.ru: one test is usually enough to understand where AI truly saves time.
erid: 2Vtzqxc7YcT
Advertisement. OOO "Iskra" INN 1685025044
Vasily Zakiev, founder of the IskraBot AI platform, lecturer at Innopolis University