Aleksandr Shulgin
LLM Developer
Python
JavaScript
TypeScript
Rust
Node JS
Computational Efficiency: 5/10
Data Pipeline & Feature Engineering: 4/10
MLOps & Deployment: 4/10
Active 18 days ago
+7 (914) 3210105 Invite to interview
Message
Download CVCV
Overview
Technical skills
Timeline
Roles
Overview
An experienced engineer focused on building production LLM-driven agents and computational research tooling at a Senior level. The strongest proven skill is engineering resilient, safety-aware LLM orchestration and automation, demonstrated by the cover_letter prompt/regeneration/guardrail logic (ai-job-search-bots/hh-autoapply/cover_letter.py) and the orchestrator/watchdog/queue patterns (ai-job-search-bots/hh-autoapply/main.py and tg_listener.py). Public code does not show any custom model training, distributed training stacks, or formal ML experiment tracking (W&B/MLflow) - those capabilities are not evidenced here.
Technical skills
Python• Senior
JavaScript
TypeScript
Rust• Junior
Node JS• Junior
Python
Requests
Aiohttp
Asyncio
Beautiful Soup
FastAPI
Rust
Cargo
Node JS
PM2
Databases
SQLite
AI/ML
NumPy
Anthropic SDK
Claude
Function Calling
Gemini
Groq
LLM
Frontend
React.js
Vite
Framer Motion
ESLint
D3.js
DevOps
GitHub Actions
CI/CD
Cloudflare
Docker
Nginx
WebSockets
Rest API
Web3
Arbitrum
Ethereum
QA
Playwright
Pytest
Timeline
AI / LLM Engineer (Python)
•
Middle
Self-employed
•
Full-Time
Built and operated production 24/7 multi-LLM pipelines for commercial and client systems, covering scoring, classification, content generation, and OCR-based translation. Designed cross-provider LLM orchestration with tool/function calling, structured outputs, quality evaluation, retries and fallbacks, plus latency-critical components in Rust. Owned end-to-end delivery including backend automation, Telegram-based control/alerts, and infra reliability on self-managed infrastructure.
Claude
Gemini
Groq
Function Calling
LLM
Asyncio
Aiohttp
FastAPI
Pytest
Playwright
Beautiful Soup
Node JS
D3.js
Rest API
WebSockets
Docker
Nginx
Cloudflare
PM2
Rust
Arbitrum
Ethereum
Senior AI/ML Engineer
Confidence: High LLM Engineer
An experienced engineer focused on building production LLM-driven agents and computational research tooling at a Senior level. The strongest proven skill is engineering resilient, safety-aware LLM orchestration and automation, demonstrated by the cover_letter prompt/regeneration/guardrail logic (ai-job-search-bots/hh-autoapply/cover_letter.py) and the orchestrator/watchdog/queue patterns (ai-job-search-bots/hh-autoapply/main.py and tg_listener.py). Public code does not show any custom model training, distributed training stacks, or formal ML experiment tracking (W&B/MLflow) - those capabilities are not evidenced here.
Model Architecture & Training
2/10
How well models are designed and trained
No custom model architectures or training loops; strong prompt engineering, LLM orchestration and regeneration loops instead of training.
Evidence
ai-job-search-bots/hh-autoapply/cover_letter.py: COVER_LETTER_PROMPT and _enforce_hidden_conditions (LLM prompt engineering, regen loop)
ai-job-search-bots/hh-autoapply/claude_client.py: ask_llm / _build_fallback_chain wrapper usage (LLM integration)
Data Pipeline & Feature Engineering
4/10
How data is prepared for models
Robust data ingestion and normalization for job sources and careful large-block numeric pipelines for sieving; checkpointing and caching used for long runs.
Evidence
ai-job-search-bots/remote-boards-aggregator/sources.py: fetch_* functions + _clean/_parse_ts normalization
erdos-computational-bounds/sieve647/sieve.py: block_tau, block_spf, checkpoint save/load, process_block (segmented sieve with vectorized operations)
Experimentation & Evaluation
3/10
How results are measured and tested
Some evaluation and verification practices (UNSAT verification, meta files, A/B styles) but no formal ML experiment tracking or reproducible hyperparameter sweeps.
Evidence
erdos-computational-bounds/sat273/export_dimacs.py: meta.json sidecar and reproducible CNF export
ai-job-search-bots/hh-autoapply/cover_letter.py: A/B styles and explicit conversion-tracking intent in comments/prompt
MLOps & Deployment
4/10
How models are shipped to production
Production-oriented agent orchestration, watchdogs, persistent state, retry/queue patterns and Telegram operator integration for deployed agents; not a full ML serving stack but solid operational engineering for LLM-driven agents.
Evidence
ai-job-search-bots/hh-autoapply/main.py: run_loop, _watchdog, cycle orchestration and checkpointing of state
ai-job-search-bots/hh-autoapply/tg_listener.py: long-polling listener, queue_action for atomic handoff to orchestrator
Computational Efficiency
5/10
How efficiently computing resources are used
Concrete computational-efficiency work: vectorized NumPy sieving, paired-divisor trick, memory-conscious DIMACS streaming and block processing for large ranges.
Evidence
erdos-computational-bounds/sieve647/sieve.py: block_tau implementation (paired-divisor scheme) and block_spf vectorized updates
erdos-computational-bounds/sat273/export_dimacs.py: streaming CNF writing to avoid holding 14*L ints in RAM
Research Depth & Innovation
4/10
Depth of research and new ideas
Solid applied research/engineering for computational mathematics and SAT encodings with independent verification; no novel ML research or new model algorithms.
Evidence
erdos-computational-bounds/sat273/encode.py: explicit CNF construction, variable mapping, independent verify_covering function
erdos-computational-bounds/sieve647/sieve.py: formal problem framing and combined detectors for two Erdos problems
Verified artifacts
Expertise
AI Agents & Agentic Workflows• Senior
LLM• Senior
Industries
Artificial Intelligence• Senior
Software• Middle
Science & Engineering• Middle
Technologies
Python• Senior
Rest API
Claude
Groq
FastAPI
WebSockets
Beautiful Soup
Function Calling
Anthropic SDK
NumPy
SQLite
Docker
Cloudflare
Gemini
Nginx
LLM
Asyncio
Requests
Aiohttp
Claude• mentioned only
Claude Code• mentioned only
Systems• mentioned only
Recommendations
- Develop production LLM agent systems that require safe prompt pipelines, guardrails and operator-in-the-loop flows (use for automation tooling, recruitment automation, or internal assistants).
- Implement and harden prompt-regeneration and hidden-condition enforcement components (expand the cover_letter style/hidden-condition framework into a reusable safety library).
- Build scalable numeric/algorithmic tooling in computational math or verification domains (segmented sieves, SAT pipelines, DIMACS export and verifier integrations).
- Integrate LLM orchestration with robust ops - monitoring, retry/backoff, state checkpoints and operator notification channels (Telegram webhook/long-poll patterns demonstrated here).
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories:
Intern Backend Developer
Confidence: Low Generalist
A junior-level generalist developer with public work that includes front-end SPA and a native desktop trading monitor, but no HUMAN-AUTHORED files were present in the designated human-authored bucket for detailed engineering scoring. The strongest provable skill from the repository text is multi-exchange market integration and API handling, as evidenced by the venues integration code that queries Binance, Bybit, OKX and others. The public material does not provide human-authored evidence for backend resilience patterns, schema migrations, transaction handling, or structured test coverage.
API Design
How well APIs are designed
Not evidenced in public code
Data Layer & Database
Working with databases
Not evidenced in public code
Scalability & Performance
Handling load and speed
Not evidenced in public code
System Architecture
Overall system structure
Not evidenced in public code
Security & Auth
Protecting data and access
Not evidenced in public code
Reliability & Observability
Stability and monitoring
Not evidenced in public code
Industries
Financial Services• Intern
Technologies
CI/CD
Cargo
Tauri• mentioned only
Recommendations
- Maintain or extend multi-exchange market-data integrations and monitoring features (alerts, caching, robust timeouts) for trading tools.
- Implement production-grade backend improvements: add structured observability (metrics, correlation ids), retries with bounded backoff, and test coverage for critical fetch/parse paths.
- Ship small-to-medium native utilities or dashboards (Tauri + React) where fast iteration and UI polish matter.
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories:
Intern Frontend Developer
Confidence: Low Generalist
A practical generalist at an early/junior level with hand-authored visual assets and infra documentation. The strongest proven skill is polished static visual design, supported by the human-authored server-hardening cover asset (Sanexxxx777/server-hardening-playbook/assets/cover.html). There is little or no human-authored evidence of UI component architecture, async state management, measured performance tuning or accessible custom widgets.
UI Component Architecture
2/10
How interface parts are built
Basic handcrafted UI/visual work (static HTML/CSS asset) is present but there is no evidence of deliberate component architecture, composition patterns or a home-grown design system in human-authored files.
Responsive & Cross-browser
1/10
Works on all screens and browsers
Little or no evidence of a responsive, cross-browser strategy in human-authored artifacts; the cover HTML uses fixed canvas/viewport sizing and no clear feature-detection or RTL/i18n readiness patterns are visible in human-authored files.
Performance Optimization
1/10
Speed of the interface
Minor attention to runtime libraries is visible via configuration, but no measured perf work, bundle analysis or virtualization strategies are present in human-authored files.
Accessibility & Semantics
1/10
Usable for everyone
Very limited accessibility evidence in human-authored files - few semantic landmarks or ARIA usage; static HTML is well-styled but not demonstrably accessible.
State Management & Data Flow
Managing data in the app
Not evidenced in public code
UX & Visual Polish
3/10
Look and feel quality
Strong visual polish for a static marketing/cover asset and clear attention to visual identity and deployment notes, but UX polish is limited to presentational artifacts rather than robust UX patterns across interactive flows.
Expertise
React• Intern
HTML & CSS• Intern
Industries
Cybersecurity• Intern
Information Technology• Intern
Technologies
JavaScript
TypeScript
Node JS• Junior
D3.js
GitHub Actions
React.js
Vite
PM2
ESLint
Framer Motion
DevOps• mentioned only
Recommendations
- Develop and own small marketing/portfolio SPA pages or brochure sites where single-file visual polish and static assets are the primary deliverable.
- Implement and maintain developer-facing documentation, deployment pipelines and operations checklists (server hardening and GitHub Actions automation).
- Contribute to UI/UX tasks that focus on isolated visual features or canvas-based mascots — avoid assigning complex SPA state or accessibility-critical components without additional evidence.
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories: