Ilyas Salimov
Full-Stack Engineer | Next.js, Node.js & AI
Ready for work
Ilyas is ready for their next project!
Agentic Financial Document Intelligence & Transaction Reconciliation Engine
• Role: Lead AI & Systems Architect
• Deliverables: Multi-Modal PDF Parser, Hybrid Agent Orchestrator, Deterministic Math Engine, Audit Trail Logger
• Timeline: 3 Weeks
OVERVIEW
Financial institutions and back-office teams spend thousands of manual hours reconciling unstructured documents (loan contracts, invoices, compliance bylaws) with relational transaction ledgers. Standard LLM wrappers fail catastrophically in this domain due to mathematical hallucinations, version conflicts, and a lack of explainable audit trails.
I engineered an agentic document intelligence system that combines multi-modal PDF extraction with deterministic validation tools to automate complex cross-referencing with 100% computational precision.
THE PROBLEM & DIAGNOSIS
LLM Math Hallucinations: Generative models cannot be trusted to perform exact financial calculations, interest compounding, or tax audits.
Unstructured vs. Tabular Asymmetry: Document facts are scattered across messy layouts, whereas transaction logs live in strict relational schemas.
Document Version Discrepancies: Clauses change across revisions; systems must resolve conflicting policies before executing reconciliation.
KEY ARCHITECTURAL DECISIONS
• Decision 01: Separation of Reasoning and Execution
The LLM acts strictly as a reasoning and extraction layer, identifying entities and formulas. All financial calculations and reconciliations are handed off to deterministic Python execution sandboxes.
• Decision 02: Cross-Source Entity Disambiguation
Built a cross-referencing engine that maps unstructured text snippets directly to SQL transaction IDs using hybrid vector-keyword scoring.
• Decision 03: Fully Auditable Chain-of-Thought
Every validation output includes a step-by-step citation trail pointing to the exact page, bounding box, and ledger entry.
MEASURABLE RESULTS & IMPACT
• 0% computational hallucination rate achieved via deterministic code execution.
• 90%+ precision in entity extraction across multi-page, complex financial agreements.
• Production-grade audit trails satisfying compliance requirements.
TECH STACK
Python 3.11, FastAPI, Pydantic, LangChain, PyMuPDF, Pandas, PostgreSQL / pgvector, Docker.
2
115
High-Performance Headless E-Commerce with Next.js BFF Architecture
• Role: Full-Stack Architect
• Deliverables: Headless Next.js Storefront, Backend-for-Frontend (BFF) Layer, WooCommerce API Adapter, Caching Strategy
• Timeline: 4 Weeks
OVERVIEW
Standard monolithic e-commerce platforms often suffer from slow page loads, theme bloat, and poor Core Web Vitals (INP/LCP), directly damaging SEO rankings and conversion rates. I architected and implemented a complete migration from a coupled WordPress/WooCommerce theme to a blazing-fast Headless Next.js storefront without losing existing customer data, order histories, or SEO juice.
THE PROBLEM & DIAGNOSIS
Monolithic Bottlenecks: PHP server rendering and plugin overhead throttled mobile page speed.
Tight Frontend/Backend Coupling: Any UI modification risked breaking checkout and core WooCommerce logic.
SEO Degradation Risk: A careless headless rewrite could break category indexing, metadata, and structured schema tags.
KEY ARCHITECTURAL DECISIONS
• Decision 01: Backend-for-Frontend (BFF) Pattern
Introduced an intermediate API layer between Next.js and WooCommerce REST endpoints. The frontend only consumes normalized, optimized payloads tailored specifically for the UI components.
• Decision 02: Hybrid Hybrid Rendering (ISR + SSG)
Implemented Incremental Static Regeneration (ISR) for high-traffic product catalog pages, ensuring instant cached load times while maintaining live stock data.
• Decision 03: Seamless Checkout & Auth Preservation
Decoupled the customer-facing frontend while preserving native WooCommerce session tokens and Stripe payment gateways.
MEASURABLE RESULTS & IMPACT
• 90+ Lighthouse Performance & SEO scores across all catalog and product pages.
• Sub-500ms initial page load times with zero downtime during backend migration.
TECH STACK
Next.js (App Router), TypeScript, WooCommerce REST API, Redis, Tailwind CSS, Docker Compose.
1
60
High-Concurrency Real-Time Logistics ERP & Dispatch Platform
• Role: Lead Full-Stack Architect
• Deliverables: Multi-Role Web Platform, Real-time SSE Dispatcher, Waybill PDF Generator, Android App API
• Timeline: 8 Weeks
OVERVIEW
A growing B2B/B2C courier and freight business was bottlenecked by manual phone dispatching, unstandardized pricing formulas, and physical paper waybills. I designed and deployed an end-to-end operations platform and mobile dispatch system serving administrators, operators, and field couriers with real-time state synchronization.
THE PROBLEM & DIAGNOSIS
Fragmented State: Dispatchers, managers, and drivers worked from different spreadsheets, leading to lost packages and miscommunication.
Ad-hoc Pricing: Freight costs were estimated manually by operators, causing revenue leakage and inconsistent quotes.
Lack of Real-Time Visibility: Order status updates depended entirely on phone calls.
KEY ARCHITECTURAL DECISIONS
• Decision 01: Strict Role-Based Access Control (RBAC)
Architected 3 distinct role interfaces (Admin, Dispatcher, Driver) on a unified database schema with row-level security and permission boundaries.
• Decision 02: Server-Sent Events (SSE) Live Feed
Replaced heavy database polling with a lightweight SSE pipeline, pushing instant status updates to dispatchers and couriers under high concurrency.
• Decision 03: Automated Pricing & PDF Waybill Engine
Built a deterministic calculation engine based on volumetric weight and zones, coupled with automated instant PDF generation.
MEASURABLE RESULTS & IMPACT
• 100% elimination of manual paper waybills via instant automated PDF generation.
• Sub-second dispatch notifications via persistent SSE connections.
• Deployed in live production powering daily logistics operations.
TECH STACK
Next.js, Express.js, Supabase, PostgreSQL, Redis, Server-Sent Events (SSE), Tailwind CSS, Docker.
1
56
Multi-Agent LinkedIn Content Orchestration & Production Pipeline
• Role: Full-Stack & AI Systems Architect
• Deliverables: Distributed Agent Pipeline, BullMQ Queue Workers, Webhook Orchestrator, React Visualizer
• Timeline: 4 Weeks
OVERVIEW
Fully automated AI publishing tools dilute brand authority by producing generic, ungrounded content. This project was built to solve the scalability bottleneck for technical founders: an autonomous 6-stage multi-agent pipeline that drafts, formats, and generates visual slide decks end-to-end while keeping a human decision-maker at the critical approval gate.
THE PROBLEM & DIAGNOSIS
Single-Prompt Fragility: A single LLM prompt cannot simultaneously handle trend extraction, technical validation, tone matching, and SVG slide generation.
The Hallucination / Tone Drift Risk: Autonomous posting without human oversight creates brand liability.
Decoupled Pipeline Failures: When one task fails in a monolithic chain, the entire run collapses without restart capabilities.
KEY ARCHITECTURAL DECISIONS
• Decision 01: Multi-Agent Queue Architecture
Engineered six specialized agents, each deployed as an isolated BullMQ worker on Redis (Researcher, Writer, Technical Reviewer, Visual Designer, Formatter, Publisher).
• Decision 02: Human-in-the-Loop (HITL) Gate
Implemented a strict state machine where the pipeline pauses at the final staging phase, providing an interactive UI for editing, approval, or regeneration before publishing.
• Decision 03: Automated SVG Carousel Generation
Created a programmatic design agent that parses technical takeaways and generates clean, dark-mode SVG/PNG carousel slides ready for social sharing.
MEASURABLE RESULTS & IMPACT
• 80% reduction in time required to produce high-authority technical slide decks.
• 100% isolation of agent failures with BullMQ automated retries and dead-letter queues.
TECH STACK
TypeScript, Next.js, Node.js, BullMQ, Redis, Groq API, Claude API, Tailwind CSS.
1
66