E.Nikitin
Избранные проекты / AI Product ManagementSelected work / AI Product Management

Проекты.Projects.

Четыре проекта: оценка стартапов, подача уведомлений, поиск работы и научный атлас. Каждый начался с конкретной задачи, которую хотелось упростить.A filing queue, a job search assistant, a research atlas, and an investment analysis prototype. Each began with a task I wanted to make easier.

01 / 04AI research prototype

A single workflow for startup assessment

The problem

Assessing an early-stage venture means reading a pitch deck, checking external sources, comparing similar companies, and accounting for both technical and business maturity. Those steps were scattered across documents and tabs.

What I built

An Adaptive Innovation Valuation Engine prototype that parses a PDF deck, retrieves relevant material, looks for comparables, applies TRL and risk adjustments, and drafts an investment-style report. The product question was how to make the reasoning traceable enough for an analyst to review.

Prototype. The report is a research aid, not an autonomous investment decision.

02 / 04Operations · AI · workflow

Turning a recurring filing task into a trackable workflow

The problem

Each organized bus trip required information from messages, spreadsheets and documents to be entered into both a DOCX notice and a multi-step government form. Multiple vehicles and drivers made manual entry especially fragile.

How it evolved

I first built local scripts for document generation and browser filling. Once batches became difficult to track, I designed Buscript around a persistent trip record, extraction of candidate fields, questions for missing details, document generation, and a queued browser worker.

The product decision

Uncertain details return to a person for confirmation. The browser worker records evidence and normally stops before the final submission. This keeps the person responsible for the filing while removing repeated entry.

The public repository contains both the original scripts and the newer application code.

View code and architecture ↗
03 / 04HR-tech · AI assistant

A job-search assistant that starts with saved vacancies

The problem

Saving a role on a phone is easy. Returning later to review requirements, compare them with a profile, and draft a relevant letter takes considerably more work.

What I built

A Chrome extension checks new saved vacancies when HH is opened. A local service fetches vacancy details, queues analysis, and sends a fit review and draft letter to Telegram. The first sync creates a baseline, so old saved jobs do not flood the queue.

The product decision

The assistant prepares material for review; it does not send applications. Candidate facts come from an editable local profile.

Local prototype. Source and setup are available on GitHub.

View code and product notes ↗
04 / 04Research · interaction design

Scientific questions as explorable maps

The problem

A complex research topic is hard to understand as a long list of papers. Readers need to see the subquestions, relationships, evidence, and places where a field remains uncertain.

What I built

A Russian-language interactive prototype with six problem maps, 72 nodes, search and filters, graph/list/material views, and detail panels that connect a node to neighboring questions and reading material.

The product decision

The maps are an editorial starting point, not a claim to classify science exhaustively. Demo materials are marked as such, and the public atlas works locally without an account.

Interactive local prototype. External source ingestion and editorial moderation are future work.

View code and product notes ↗