ML ENGINEER / DATA SCIENTIST

Gonchar Daniil

I work on tasks of any complexity and domain

5+ years
NLP · LLM · RAG

01 / Expertise

Areas of expertise

Corporate RAG systems for search across Confluence, DWH, SQL data marts, PDFs and scans. I separate the offline knowledge pipeline from the online answer flow.

Hybrid retrievalLLM rerankingVersion-aware retrievalContext tighteningSource attribution

I control versions and access, combine dense and sparse retrieval, narrow a broad candidate pool through LLM reranking, and generate answers only from compact context with mandatory source attribution.

Stack

bge-m3 + BM25Qdrant · HNSWQwen3-NextStructure-aware chunkingSchema-aware promptingDocling + EasyOCR

Tool orchestrationStateful workflowsDependency graphsSafe executionEarly stop

NERLayout parsingOCRDoclingEasyOCR

Model servingStreaming pipelinesInference optimizationExperiment trackingDrift monitoring

02 / Projects

The approach, told through real problems

Case 01

CorpGPT Knowledge RAG

Problem: Search across Confluence, DWH, SQL, tables and scans.
Solution: Hybrid retrieval, Qdrant HNSW, Docling + EasyOCR, rerank, source attribution and version-aware filters.
Role: Design of the retrieval pipeline and generation constraints.

Status: Details on request.

Case 02

Data Change Analysis Agent

Problem: Explain changes in data marts and calculations.
Solution: SQL and regulation parsing, dependency graph, ReAct and tool-calling; simulation instead of direct SQL execution.
Role: Agent architecture and protective constraints.

Status: Details on request.

Case 03

Data Incident RCA Agent

Problem: Find root causes of data degradations across ETL, releases, SLA and lineage.
Solution: Stateful workflow, incident context, hypothesis prioritisation and early-stop.
Role: Reasoning pipeline and search depth control.

Status: Details on request.

Case 04

Document Intelligence Pipeline

Problem: Extract data from unstable PDFs, tables and scans.
Solution: Layout parsing, OCR, regex, NER and normalisation.
Role: End-to-end document processing.

Status: Details on request.

03 / Testimonials

Words from colleagues and partners

Testimonials from colleagues and partners will appear here once publication is approved.

04 / Contact

Have an ML, NLP, or LLM challenge? Let's talk.

Telegram @gonchardaniil GitHub Lainterus1

Gonchar Daniil

Areas of expertise

LLM & RAG / CorpGPT

AI Agents

NLP & Document Intelligence

Production ML

The approach, told through real problems

CorpGPT Knowledge RAG

Data Change Analysis Agent

Data Incident RCA Agent

Document Intelligence Pipeline

Words from colleagues and partners

Leave a testimonial

Have an ML, NLP, or LLM challenge? Let's talk.