Python, FastAPI, React Native, LangGraph, pgvector, Supabase, OpenAI Built a full-stack application (FastAPI backend, Expo mobile app) that translates abstract emotional input into personalised book and film recommendation using a multi-agent LangGraph pipeline Architected a parallel retrieval system combining pgvector cosine similarity search over a 6,000+ work corpus (Supabase) with live Wikidata SPARQL queries, Reddit/ web context via Tavily, and GPT synthesis Implemented Supabase Auth with JWT middleware, per-user daily usage limits, and session history Diagnosed and reduced pipeline latency from 131s to 75s using LangSmith observability tracing, identifying bottlenecks

Python, FastAPI, React, Supabase, pgvector, Railway Built a full-stack regulatory intelligence platform monitoring EU, Austrian, German and Swiss tax law across 5 APIs with automated daily digests generated by Claude Architected a RAG pipeline using pgvector semantic search, multi-documental retrieval and streaming LLM responses across a jurisdiction-filtered document corpus Deployed to Railway as a live demo with authentication, React frontend served via FastAPI static files

Python, FastAPI, LangGraph, pgvector, Supabase, OpenAI, LangSmith Built an agentic RAG pipeline over 31,100 KJV Bible verses and 100 story summaries with modern framing, embedded via OpenAI text-embedding-3-small into Supabase pgvector Designed a LangGraph StateGraph agent with query expansion, self-evaluating retrieval quality, and iterative query reformulation Built server-side session memory enabling multi-turn dialogue - deep dives, alternative story candidates and follow up questions

Python, FastAPI, ChromaDB, spaCy, BookNLP, Claude API Built a spoiler-free RAG reading companion with strict page-boundary constraints enforced at the vector DB query layer, preventing spoilers Designed a 3-tier NER pipeline (Supabase entity registry -> BookNLP transformer -> spaCy + fallback) for robust character extraction Developed a scene roleplay system enabling multi-turn dialogue with book characters, grounding responses in built character profiles

Valued over 40+ loan portfolios using DCF and NPV approaches in the context of forward flow sale programs Negotiated valuation methods with counterparties to achieve target IRR threshold Automated cash flow models using VBA, significantly reducing manual processing time

Identified a business opportunity that generated up to 70% of the company’s profit Secured 50+ deals with international manufacturers, reducing company expenses by 20% Negotiated to list the products on the leading marketplace Drop which increased exposure by 270%

Analyzed key financial results, performance, budget, forecast, KPIs and trends of CE & EE Corporates business with a specific focus on the Product Factories and segments Supported the dialogue with Product Factories and local Banks to gather and elaborate information regarding the financial results achieved

After a careful study of currency and AML legislation I monitored over 250 daily transactions to identify those subject to regulatory control. I then submitted the corresponding request to an NCA for further review

I requested information about the clients to evaluate credit risk levels by compiling this information and getting the final score according to bank rules and guidelines. If a client had a collateral property I would work closely with a team to estimate its value using various numerical methods such as the Market Method or the D&A method.

On average I prepared documents for 25 interbank transactions a day and every 4 hours I set the exchange rates for the Treasury. Prepared documents and information for subsequent exchange with the Federal Customs Service of Russia.