Daniil Pupshev - Python Developer

Overview

A data-engineering focused practitioner (senior level) building end-to-end video analytics and operational reporting pipelines. The strongest proven skill is delivering production-focused video-processing and reporting - including OCR-based video time-anchor resolution and a comprehensive Excel KPI report generator - that connects CV outputs to business metrics. There is limited evidence of formal statistical analysis, model training pipelines, automated ML evaluation suites or pinned environment/data-versioning in public code.

Phone

Timeline

Lab programmer · Middle

EKRAN LLC · Full-Time

May 2025 to Present 1 Year 6 Months Zhukovsky In office

Develops internal automation tools and Python backend services, working with REST APIs and data storage using PostgreSQL and SQLite. Implements the full task cycle: clarifies requirements, designs solutions and data models, builds and validates functionality in real scenarios, and supports production usage. Uses Git-based version control and CI/CD workflows, along with SSH-based operations and maintenance.

Rest API

PostgreSQL

SQLite

Git

CI/CD

Moscow Polytechnic University

Bachelor's Degree · Programming and Information Technologies in Dynamics and Strength

2028 Moscow, Russia

Projects

CalcAUO Attenuation Calculation System

Developed a multi-user application to automate engineering calculations using a client-server architecture. Implemented a FastAPI-based REST API, database access through SQLAlchemy, and Pydantic-based validation for inputs and outputs. Added calculation history storage and retrieval; the app is used by a team and reduces test result processing and reconciliation time from several hours to under half an hour.

Project_1 Video Analytics System

Designed and implemented a video analytics pipeline that registers videos, links timestamps via OCR, detects and tracks visitors, and matches events with receipt data. Built the data storage and processing structure in PostgreSQL and implemented sequential processing for large data volumes. Optimized throughput to process about 24 hours of 25 GB video in roughly 1.5–2 hours and generate daily analytics summaries for marketing and analytics use.

Technical skills

Languages

2

Python

SQL

Python

10

Requests

Pydantic

SQLAlchemy

FastAPI

Click

Asyncio

aiogram

Alembic

Uvicorn

PySide6

Databases

3

PostgreSQL

SQLite

pgvector

AI/ML

6

NumPy

OpenCV

Pandas

YOLO

PaddlePaddle

LLM

DevOps

3

Git

Rest API

CI/CD

Other

9

OpenAI

Deep Learning

GitHub

Linux

Docker

OpenAI Codex

RAG

Prompt Engineering

OCR

Senior AI/ML Engineer Confidence: Medium ML Engineer

Computer vision ML engineer (senior-level) focused on applied video analytics for retail cashier-zone workflows with strong pipeline and domain-algorithm engineering. The strongest proven skill is building robust video-to-analytics pipelines and diagnostic tooling as shown by the time-anchor resolution code (pechvision/video/time_anchor.py) and end-to-end processing CLI (pechvision/cli/main.py). Public code does not show custom model training, experiment tracking, or GPU/quantization efficiency engineering, and there is limited evidence of automated tests or CI for ML experiments.

Model Architecture & Training

2/10

How well models are designed and trained

Model usage and integration are solid (YOLO, InsightFace, PaddleOCR) but there are no custom training loops, schedules, or model architecture design in the code.

Evidence

pechvision/vision/person_detector.py: imports and uses ultralytics YOLO

pechvision/vision/faces.py: build_insightface_app and embedding normalization (InsightFace integration)

Data Pipeline & Feature Engineering

5/10

How data is prepared for models

Substantial, domain-specific data pipelines and feature engineering for video timelines, visit/session aggregation and receipt normalization are implemented and show careful edge-case handling.

Evidence

pechvision/video/time_anchor.py: OCR anchor collection, grouping and representative selection algorithms

pechvision/visits/session_builder.py: grouping visits into session candidates and session key building

pechvision/receipts/reader.py: receipts parsing and normalization logic

Experimentation & Evaluation

2/10

How results are measured and tested

Reporting and metric calculation exist, but there is no experiment tracking, reproducible experiment configs, A/B setup, or evaluation harness for model experiments.

Evidence

pechvision/reports/excel_export.py: builds daily and total metrics for reporting

pyproject.toml: pytest configured under dev dependencies but no visible experiment tracking (W&B/MLflow) or tests present

MLOps & Deployment

4/10

How models are shipped to production

Good operational engineering: CLI entrypoints, DB migrations, processing-run lifecycle and config snapshotting; suitable for production pipelines but not a full MLOps stack with CI/CD or monitoring shown.

Evidence

pechvision/cli/main.py: CLI commands, progress reporting and processing run lifecycle handling

alembic/versions/*: alembic migrations present (DB lifecycle)

pechvision/db/session.py: engine and session factory creation

Computational Efficiency

2/10

How efficiently computing resources are used

Some practical efficiency considerations (frame stepping, limits, progress and speed reporting) but no GPU memory/throughput profiling, batching, model quantization, or distributed inference code.

Evidence

pechvision/video/frames.py: iter_video_frames with frame_step parameter

pechvision/cli/main.py: progress speed calculation and inference frame bookkeeping

Research Depth & Innovation

4/10

Depth of research and new ideas

Shows domain-specific algorithmic work (time-anchor grouping/selection, identity quality scoring), indicating applied research/engineering but not novel published research or paper-level contributions.

Evidence

pechvision/video/time_anchor.py: find_largest_consistent_attempt_group and select_representative_attempt implementations

pechvision/vision/faces.py: calculate_identity_quality_metrics with multi-factor scoring

Expertise

Computer Vision & Image Analysis· Senior

Document Intelligence & OCR· Middle

Industries

Data & Analytics· Middle

Technologies

SQL

pgvector

Prompt Engineering

CI/CD· since 2025

Git· since 2025

SQLite· since 2025

PaddlePaddle

LLM

RAG

PySide6

OpenAI

OpenAI Codex

GitHub

OCR

Linux

Recommendations

Develop customer-facing video analytics features and production pipelines that detect/track people, extract faces, and produce aggregated reports (use the video pipeline, time_anchor and reports modules).
Integrate and harden OCR and face recognition components for deployment (work on pechvision/video/ocr.py and pechvision/vision/faces.py to add robustness, batching and fallback strategies).
Implement experiment tracking and evaluation harness (W&B or MLflow) and add reproducible test/benchmark suites for model inference and pipeline metrics.
Work on MLOps tasks like inference performance profiling, model quantization/optimization and CI/CD for model and pipeline deployments.

Repositories

The developer's experience in this domain has been verified based on AI analysis of the following repositories:

Project_1 CalcAUO RecNumbers

Senior Data Scientist Confidence: Medium Data Engineer

A data-engineering focused practitioner (senior level) building end-to-end video analytics and operational reporting pipelines. The strongest proven skill is delivering production-focused video-processing and reporting - including OCR-based video time-anchor resolution and a comprehensive Excel KPI report generator - that connects CV outputs to business metrics. There is limited evidence of formal statistical analysis, model training pipelines, automated ML evaluation suites or pinned environment/data-versioning in public code.

Statistical Rigor

2/10

Correct use of statistics

Minimal statistical rigor; no hypothesis testing, uncertainty quantification or causal analysis. Code includes aggregations and medians but lacks formal statistical testing or explicit treatment of uncertainty.

Evidence

pechvision/reports/excel_export.py:calculate_daily_metrics

pechvision/video/time_anchor.py:find_largest_consistent_attempt_group

Data Wrangling & Cleaning

6/10

Preparing and cleaning data

Strong data wrangling and provenance work with parsers, normalizers, importers and sessionization logic; careful handling of timestamps and format normalization is evident.

Evidence

pechvision/receipts/reader.py:normalize_receipts_file

pechvision/receipts/importer.py:import_receipts

pechvision/visits/session_builder.py:group_visits_into_session_candidates

Exploratory Analysis & Visualization

6/10

Exploring and visualizing data

Practical, purposeful visualization and diagnostics aimed at operational monitoring and business reporting rather than exploratory research. Excel reports include KPI tables, charts and written notes tying metrics to quality assessments.

Evidence

pechvision/reports/excel_export.py:build_general_sheet

pechvision/cli/main.py:timeline_check_command

Predictive Modeling

6/10

Building models that predict

Applied predictive/ML inference engineering for CV: model integration, embedding normalization and identity-quality scoring are implemented with awareness of practical failure modes, but no training pipelines or rigorous model evaluation harness are present.

Evidence

pechvision/vision/faces.py:calculate_identity_quality_metrics

pechvision/vision/person_detector.py:get_person_detector

Business Insight & Impact

5/10

Turning analysis into business value

Analysis is tied to operational business metrics and reporting; the code produces actionable Excel reports with KPIs and coverage metrics, but there is limited explicit business-impact modelling (cost of errors, A/B style experiment framing).

Evidence

pechvision/reports/excel_export.py:build_general_sheet

pechvision/cli/main.py:export_report_command

Reproducibility & Notebook Hygiene

5/10

Clean, repeatable analysis

Reasonable reproducibility practices: config loader and snapshotting, DB session factories and alembic migrations are present. There is no explicit pinned environment or data/versioning tooling in the code shown.

Evidence

pechvision/config/snapshot.py:build_config_snapshot

alembic/versions/21f5b49cfe44_create_initial_schema.py:upgrade

Expertise

Data Science· Middle

Industries

Commerce· Senior

Science & Engineering· Middle

Transportation & Logistics· Middle

Technologies

Deep Learning

PostgreSQL· since 2025

OpenCV

YOLO

Pandas

NumPy

Recommendations

Develop and productionize robust inference pipelines for video analytics including containerized model serving and resource-aware batching for GPU/CPU inference.
Add automated model evaluation and monitoring: unit tests for CV components, offline evaluation harness, and drift/accuracy dashboards for embeddings and OCR.
Extend reproducibility by adding environment pinning, requirements lockfiles, and dataset/version tracking (DVC or similar) plus CI checks for migrations and core pipeline regressions.

Repositories

The developer's experience in this domain has been verified based on AI analysis of the following repositories:

Project_1 CalcAUO RecNumbers

Senior Backend Developer Confidence: High API Engineer

Backend engineer (Senior) focused on building domain-specific analytic pipelines and data models with strong practical knowledge of local video analytics and reporting. The strongest proven skill is designing and evolving a production-grade relational schema with vector search support and migrations, evidenced by pechvision/db/models.py and multiple Alembic revisions (alembic/versions/*). Public code lacks automated test suites, CI/CD manifests and production observability (structured logs/metrics) as explicit artifacts.

API Design

4/10

How well APIs are designed

API design shows a working REST surface with token-based header auth and pydantic validation but lacks explicit versioning, formal idempotency strategy, or advanced pagination/error contract patterns.

Evidence

CalcAUO/server/api/routes/notes.py: _extract_token and route handlers implementing header token auth

CalcAUO/server/api/schemas.py: Pydantic DTOs and field validators used for request/response validation

CalcAUO/server/api/routes/auth.py: login/logout routes and use of auth service

Data Layer & Database

6/10

Working with databases

Data layer is well designed with SQLAlchemy declarative models, explicit indices and constraints, pgvector usage and an Alembic migration chain showing schema evolution and thoughtful indexing for vector search.

Evidence

pechvision/db/models.py: detailed SQLAlchemy models, indices and VECTOR(512) columns with a HNSW index

Project_1/alembic/versions/897039a988a4_add_pgvector_embeddings.py: migration adding pgvector embeddings

pechvision/db/session.py: make_engine and make_session_factory patterns and session management

Scalability & Performance

4/10

Handling load and speed

Some performance and scaling attention (vector index, cached model init, reusable VideoCapture) but no evidence of distributed processing, queue-based decoupling or documented load-tested optimizations.

Evidence

pechvision/vision/faces.py: @lru_cache get_insightface_app to avoid repeated expensive initialization

pechvision/video/time_anchor.py: collect_ocr_anchor_attempts reuses a single cv2.VideoCapture for multiple OCR attempts

pechvision/db/models.py: vector index (HNSW) for embedding search

System Architecture

5/10

Overall system structure

Clear modular decomposition and separation of concerns across config, video pipeline, identity and DB layers with CLI orchestration; architecture is monolithic but deliberate and well-structured for a single-service analytic pipeline.

Evidence

pechvision/cli/main.py: command group orchestrating video registration, pipeline runs and persistence

pechvision/config/schema.py: typed configuration schema separating project, DB, video, detection and faces concerns

pechvision/video/pipeline.py: pipeline orchestration and use of Timeline/VisitsBuilder components

Security & Auth

4/10

Protecting data and access

Basic auth and input-safety measures are present (token issuance, header extraction, pydantic validators); however there is limited evidence of token lifecycle management (refresh/revocation), dependency auditing, or explicit protections against SSRF/SQ injection beyond using SQLAlchemy ORM.

Evidence

CalcAUO/server/services/auth_service.py: token issuance and hashing logic

CalcAUO/server/api/routes/notes.py: header token extraction and resolve_user_id usage

pechvision/config/schema.py: Pydantic-based config validation for many input boundaries

Reliability & Observability

4/10

Stability and monitoring

Code includes defensive error handling, session rollback/commit patterns and resource cleanup but lacks structured logging, metrics/alerts, or richer resilience patterns like retries with jitter or circuit breakers.

Evidence

pechvision/cli/main.py: try/except with session.rollback/commit, progress updates and finalization

pechvision/video/time_anchor.py: VideoCapture opened/released in try/finally to avoid resource leaks

TG_bot_RN/main.py: polling loop with restart-on-exception for basic runtime resilience

Expertise

Python· Middle

Databases & Vector Storage· Middle

Microservices & API Architecture· Middle

Messaging & Real-time· Middle

Industries

Commerce· Middle

Science & Engineering· Middle

Technologies

Python· since 2025 · Senior

Rest API· since 2025

SQLAlchemy

FastAPI

Asyncio

Pydantic

Uvicorn

Requests

Click

Alembic

aiogram

Recommendations

Develop end-to-end video analytics pipelines and integrations where model inference, timeline calibration and DB persistence are central.
Implement backend REST APIs and reporting endpoints that expose aggregated analytics and run controlled ingest/upsert transactions for visit/receipt linkage.
Own Postgres schema evolution and vector search features (pgvector, HNSW index) and optimize query/index decisions for analytics workloads.

Repositories

The developer's experience in this domain has been verified based on AI analysis of the following repositories:

Project_1 CalcAUO TG_bot_RN

Job preferences

Work permits: Russia
Citizenship: Russia
Employment type: Full-Time