TIM TUEV / TASHKENT, UTC+5 / OPEN TO REMOTE & RELOCATION

AI Deployment
Strategist

Bridging business, product and AI engineering.

I turn business and product problems into AI systems — from model selection and inference strategy to agents, memory, retrieval and deployment.

[ DISCUSS A ROLE ] CV GITHUB ↗ LINKEDIN ↗
SYNTHETIC MIND // TOPOLOGY GAZE SYNTHETIC MIND
AVAILABLE FOR HIRE AVAILABLE
INITIALIZING 3D NEURAL TOPOLOGY...
┌
┐
└
┘
GAZE: Y: 0.0° // P: 0.0°
◉ TRACKING
CORE POSITIONING

From Business Problem
to AI System

BUSINESS

INPUT →
  • Problem Framing
  • User Needs & Workflows
  • Business Value
  • Regulatory Constraints
  • Cost & Unit Economics

AI SYSTEM

THE MIDDLE
  • Model Selection & Grounding
  • LoRA / Domain Adaptation
  • Inference & Quantization
  • Agents & Tool Calling
  • Memory & Hierarchical Retrieval
  • Rigorous Evals & Benchmarks

PRODUCT

→ OUTPUT
  • System Architecture
  • Working Prototypes
  • Deployment & Packaging
  • Non-Hallucinatory UX
  • KPI Measurement
"My work happens in the middle."

Business teams often know exactly what needs to improve but not which AI approach makes technical or economic sense. AI engineers understand the underlying models and infrastructure but may not own the product, user, or business context. I connect the two.

DEPLOYED PRODUCTION ENGINES

Selected Systems

Autonomous project operations, compliance boundaries, and candidate intelligence pipelines built for production workloads.

SYSTEM / 003 AI OPERATIONS & AGENTS ACTIVE
2026

Philipp — Ambient AI Project Manager

An ambient AI project manager that listens to team chatter without conversational spam, maintains a structured Kanban board, and dispatches validated tasks to an autonomous coding execution loop.

0 API Tokens
DETERMINISTIC SYSTEMD CRONS
Issue → PR
AUTONOMOUS CODING EXECUTION LOOP
[ AGENTIC DISPATCH ]

A 12-second rolling debounce batch evaluates PM vs Speech decisions in a single forward pass. Approved tickets invoke headless Antigravity CLI workers to checkout branches, synthesize code, run tests, and open PRs.

[ CASE STUDY & ARCHITECTURE ] SYSTEM SPECS →
TOPOLOGY: AMBIENT PM & AUTONOMOUS CODE WORKER 
┌───────────────────────────────────────────────┐
│ CONVERSATION STREAM (Telegram Groups/DMs)     │
└───────────────────────┬───────────────────────┘
                        │ 12s Debounced Conversation Batch
                        ▼
┌───────────────────────────────────────────────┐
│ GEMINI 3.8 FLASH (Single-Pass Dual Decision)  │
│  ├─ PM Decision: Triage vs Candidate Inbox    │
│  └─ Speech Decision: Clarify vs [NO_REPLY]    │
└───────────┬───────────────────────┬───────────┘
            │ Mutation              │ Output
            ▼                       ▼
┌─────────────────────────┐ ┌───────────────────┐
│ SQLite/WAL Task Store   │ │ Telegram Suppress │
│ Mechanical Dedupe Check │ │ Silent Bot Filter │
└───────────┬─────────────┘ └───────────────────┘
            │ Dispatched Ticket
            ▼
┌───────────────────────────────────────────────┐
│ HEADLESS ANTIGRAVITY WORKER (Autonomous Loop) │
│ - Branch checkout (task/KEY)                  │
│ - Code synthesis, unit test execution & diff  │
│ - Generates pull request & reports diffstat   │
└───────────────────────────────────────────────┘
PYTHON 3.12 / GEMINI 3.8 FLASH / SYSTEMD TIMERS
SYSTEM / 004 TALENT INTELLIGENCE & EVALUATION ACTIVE
2026

HR Screener

An asynchronous candidate screening platform that scrubs Russian PII and conversational noise locally before cloud transit, yielding ~35% token compression and grounded competency evaluations.

PII Masked Before Cloud
CLIENT-SIDE REGEX & LEXICAL SANITIZER
< 8s / ~$0.001
ASYNC WORKER (DEEPSEEK V4 FLASH)
[ PRIVACY & COMPLIANCE ]

Inbound MyMeet transcripts are verified via HMAC-SHA256, stripped of timestamps and speech fillers, and anonymized into [PERSON_N] tokens before cloud dispatch, with micro-cost auditing down to fractional cents.

[ CASE STUDY & ARCHITECTURE ] PIPELINE SPECS →
TOPOLOGY: DETERMINISTIC SANITIZATION & EVALUATION 
┌───────────────────────────────────────────────┐
│ INCOMING TRANSCRIPT (MyMeet Webhook / DOCX)   │
└───────────────────────┬───────────────────────┘
                        │ HMAC-SHA256 & Idempotency Key
                        ▼
┌───────────────────────────────────────────────┐
│ GO ECHO API GATEWAY (`cmd/api`)               │
│ - Constant-time signature verification        │
│ - Enqueue tasks.TypeEvaluateTranscript        │
│ - Immediate HTTP 202 Accepted response        │
└───────────────────────┬───────────────────────┘
                        │ Redis Task Queue
                        ▼
┌───────────────────────────────────────────────┐
│ ASYNQ TASK WORKER ENGINE (`cmd/worker`)       │
│  ├─ Strip MyMeet timestamps & speech fillers  │
│  ├─ Mask Cyrillic names to [PERSON_N] tokens  │
│  ├─ Structure into XML <dialogue> (-35% size) │
│  └─ DeepSeek V4 Flash: JSON competency score  │
└───────────────────────┬───────────────────────┘
                        │
                        ▼
┌───────────────────────────────────────────────┐
│ POSTGRESQL AUDIT TRAIL & TOKEN LEDGER         │
│ - Strict separation of raw text vs scrubbed   │
│ - Exact prompt/completion cost accounting     │
└───────────────────────────────────────────────┘
GO 1.21+ / ECHO V4 / ASYNQ / REDIS / POSTGRESQL
[ EXPLORE ALL 5 SYSTEMS IN DIRECTORY → ]
SYSTEMIC PERSPECTIVE

Across the AI Stack

The layers I work across — from model selection to product adoption.

01 MODELS
model selection capabilities mapping reasoning architectures multimodality dense vs MoE
capability maps over hype
02 ADAPTATION
synthetic datasets LoRA fine-tuning alignment & steering evaluation harnesses
domain data beats bigger models
03 INFERENCE
CUDA & MLX quantization (AWQ/GGUF/EXL2) llama.cpp / vLLM hardware trade-offs
latency & cost are architecture decisions
04 SYSTEMS
agents & tool use MCP servers hierarchical retrieval memory substrates provenance
memory & retrieval define the ceiling
05 DEPLOYMENT
latency budgets cost & token economics privacy & air-gapped reliability local vs cloud
constraints first: privacy, budget, reliability
06 PRODUCT
non-hallucinatory UX workflow integration user adoption business value system design
adoption is the only benchmark
Tim Tuev .01 Tashkent, UZ
The goal is to choose the simplest system that solves the hardest problem
simple
simple
simple less, but sharper »
HANDS-ON LABORATORY LOGS

Inference Lab

Running models, not just calling APIs. Hands-on experiments with open models, inference runtimes, hardware limits, and deployment trade-offs.

EXP / 001 2026

Qwen 3.8 27B / RTX 3090 / Vast.ai

RUNTIMES vLLM → llama.cpp
TESTED MTP Speculative, Quantized Inference (Q4_K_M), Remote GPU Deployment
DECISION: llama.cpp
WHY: Significantly lower operational friction, stable memory footprint under 24GB VRAM, and predictable cold-start recovery for ad-hoc batch processing.
EXP / 002 2026

MLX / Apple Silicon Unified Memory

RUNTIMES MLX Swift / Python C++ Binding
TESTED 4-bit vs 8-bit KV Cache, Zero-Copy Prompt Prefill on M-series
DECISION: 4-bit Quantized KV
WHY: 2.4x throughput preservation with unnoticeable degradation in retrieval context needle extraction.
EXP / 003 2026

MoE Expert Offloading Latency

SETUP Mixtral / DeepSeek-V2 hybrid CPU-GPU
TESTED PCIe bandwidth bottleneck vs active expert memory pin
DECISION: Pin Router in VRAM
WHY: Routing layers must never leave VRAM; dynamic expert streaming over PCIe 4.0 degrades tokens/sec by 65%.
VIEW ALL EXPERIMENTS & BENCHMARK LOGS →
RIGOROUS EVIDENCE

Field Notes

Published research notes, benchmark reconstructions, and empirical investigations.

RESEARCH NOTE / 16 AUG 2026 / LONGMEMEVAL / SKINKI

When Graph Retrieval Fails to Transfer

A typed relation graph raised synthetic multi-hop recall@10 from 0.325 to 0.800. On a small LongMemEval sample, it fell below BM25. An early research note on benchmark-shaped wins, real dialogue, and what failed next.

0.325 → 0.800
SYNTHETIC MULTI-HOP RECALL@10
0.168 vs 0.193
LONGMEMEVAL: GRAPH VS BM25
AI MEMORY RETRIEVAL GRAPHRAG SKINKI
READ FULL NOTE →
VIEW ALL RESEARCH ARTICLES & METHODOLOGIES →
FOUNDATIONAL BACKGROUND

The Other Half

Before working deeply with AI systems, I spent nearly a decade in design, product and creative leadership.

That background became the other half of my current work: understanding users, business constraints, communication, information systems and product experience. It is why I don't build tech for tech's sake.

01
BRAND & DESIGN SYSTEMS
Multi-layered typography, design tokens, identity systems, and cohesive digital design languages.
02
INFORMATION & NAVIGATION SCHEMES
Complex information architectures, wayfinding systems, and hierarchical data visualization.
03
PRODUCT & CUSTOMER JOURNEYS
Zero-to-one product design, user onboarding, high-conversion funnels, and enterprise UX.
PROFILE

About

TIM TUEV / 2026 TASHKENT
Portrait of Tim Tuev
TIM TUEV
AI Deployment Strategist
FOCUS AI Systems Architecture & R&D
ORIGIN Product & Creative Leadership
LOCATION Tashkent, UZ (UTC+5)
STATUS Ready for International Roles

The recent AI wave pulled me from creative leadership into models, inference, agents and AI systems R&D.

Today my strongest skill sits between two worlds: understanding what a business needs, understanding what current AI technology can actually do, and designing the system that connects them.

I work heavily with coding agents for implementation and rapid iteration. My ownership is in problem framing, architecture, experiments, evaluation and product decisions, while I continue developing deeper independent engineering fundamentals.

HIRING ENGAGEMENT

Currently Open To

International remote teams, roles with relocation potential, and selected high-impact opportunities in Tashkent.

01
AI DEPLOYMENT
Translating enterprise workflows into practical, cost-effective AI solutions.
02
AI SOLUTIONS ARCHITECTURE
End-to-end inference, retrieval, agentic & local-first memory stacks.
03
TECHNICAL AI PRODUCT
Business KPIs, technical teams and non-hallucinatory UX in one owner.
04
FORWARD DEPLOYMENT
Hands-on client integration, rapid prototyping, and engineering iteration.
BASED IN Tashkent, Uzbekistan
TIMEZONE UTC+5 (Flexible overlap)
WORK MODE Remote / Relocation / Local
LANGUAGES English — Professional Working Proficiency / Russian — Native
[ DISCUSS A ROLE — TIMOTHY.TUEV@GMAIL.COM ]
TELEGRAM ↗ / LINKEDIN ↗ / VIEW CV →