Green Leasing RAG Assistant
A full-stack internal support tool that helps leasing teams handle electric vehicle, charging infrastructure, and green mobility cases with source-grounded answers. The assistant combines hybrid retrieval, real-time streaming, compliance guardrails, citation tracking, and case history in a focused employee workspace. It supports practical checklists, missing-fact analysis, caveats, and draft client replies without making binding credit, legal, tax, subsidy, or taxonomy decisions.
React
TypeScript
FastAPI
OpenAI
Docker
About the project
The Product: An internal RAG assistant designed for employees who support green leasing and e-mobility cases. The application helps sales, service, fleet, SME, ESG, compliance, and risk teams prepare first responses for electric vehicles, wallboxes, charging stations, photovoltaic systems, battery storage, and combined fleet scenarios.
What it Does: It turns a raw client request into a structured advisory response with a short answer, assumptions, missing facts, document checklist, next steps, caveats, cited sources, and a conservative draft client reply. Instead of forcing employees to search through product pages, policy notes, regulatory references, and synthetic demo FAQs separately, the assistant surfaces the relevant evidence path inside one case workspace.
How it Works: Built on a modern full-stack architecture (React/Vite + Python/FastAPI), the system indexes approved documentation into a knowledge base and retrieves evidence through lexical BM25 search with optional dense Faiss retrieval. Topic-aware query expansion helps route operational, regulatory, subsidy, tax, reimbursement, GDPR, and green-claims questions toward the right source families before the LLM generates an answer.
The Advantage: The platform is engineered for safe employee assistance rather than automated decisioning. A scope guard keeps conversations focused on green leasing workflows, every answer is expected to cite supporting sources, and time-sensitive subsidy, tax, and funding statements are framed as requiring current-source verification. If model-backed generation or dense retrieval is unavailable, deterministic fallbacks keep the demo usable while preserving the same cautious answer structure.
Learn tech information for this project
About the project
The Product: An internal RAG assistant designed for employees who support green leasing and e-mobility cases. The application helps sales, service, fleet, SME, ESG, compliance, and risk teams prepare first responses for electric vehicles, wallboxes, charging stations, photovoltaic systems, battery storage, and combined fleet scenarios.
What it Does: It turns a raw client request into a structured advisory response with a short answer, assumptions, missing facts, document checklist, next steps, caveats, cited sources, and a conservative draft client reply. Instead of forcing employees to search through product pages, policy notes, regulatory references, and synthetic demo FAQs separately, the assistant surfaces the relevant evidence path inside one case workspace.
How it Works: Built on a modern full-stack architecture (React/Vite + Python/FastAPI), the system indexes approved documentation into a knowledge base and retrieves evidence through lexical BM25 search with optional dense Faiss retrieval. Topic-aware query expansion helps route operational, regulatory, subsidy, tax, reimbursement, GDPR, and green-claims questions toward the right source families before the LLM generates an answer.
The Advantage: The platform is engineered for safe employee assistance rather than automated decisioning. A scope guard keeps conversations focused on green leasing workflows, every answer is expected to cite supporting sources, and time-sensitive subsidy, tax, and funding statements are framed as requiring current-source verification. If model-backed generation or dense retrieval is unavailable, deterministic fallbacks keep the demo usable while preserving the same cautious answer structure.
Learn tech information for this project
Green Leasing RAG Assistant
Internal AI workspace for source-grounded green leasing answers, document checklists, compliance caveats, and cited client replies.
ReactTypeScript
PythonFastAPIOpenAIGreen Leasing RAG Assistant
Internal AI workspace for source-grounded green leasing answers, document checklists, compliance caveats, and cited client replies.
ReactTypeScriptPythonFastAPIOpenAI
Features
Source-Grounded Answers
The assistant is built around evidence-backed responses for green leasing workflows. A user can ask about EV eligibility, SME fleet replacement, wallbox packages, ESG taxonomy checks, subsidy caveats, tax reimbursement, or small-business document requests, and the system answers in a predictable structure that leasing employees can review and reuse.
Every generated answer is grounded in approved knowledge-base sources and includes citations that can be inspected in the sources panel. The frontend normalizes answer sections into a compact table, highlights citations inline, and keeps the cited evidence available for follow-up questions so employees can trace how a recommendation was formed.
Hybrid Retrieval Pipeline
The backend combines document ingestion, chunking, source inventory metadata, lexical BM25 retrieval, and optional dense Faiss search. Topic profiles expand queries with domain-specific terms for operational leasing processes, charging infrastructure, EU Taxonomy, green claims, subsidies, reimbursement, tax, and GDPR-sensitive fleet data.
This retrieval layer helps the assistant distinguish between product/process guidance, regulatory criteria, and synthetic demo examples. It also preserves source title, URL, retrieval date, freshness status, section metadata, and review hints, giving employees a practical route from conversational output back to the underlying evidence.
Guardrailed Case Workspace
The user experience is designed as a case workspace rather than a generic chatbot. Recent cases are stored locally, the conversation panel streams the assistant response in real time, and the sources panel collects evidence used across answers. Suggested prompts guide employees toward high-value workflows such as EV fleet assessment, charging infrastructure documents, subsidy caveats, and client reply drafting.
Domain guardrails keep the product aligned with internal support boundaries. The assistant can prepare checklists, missing-fact questions, caveats, and draft communications, but it does not approve credit, provide final legal or tax advice, guarantee subsidies, or make final taxonomy-alignment claims. That restraint makes the tool more suitable for regulated leasing operations and internal review workflows.
Features
Source-Grounded Answers
The assistant is built around evidence-backed responses for green leasing workflows. A user can ask about EV eligibility, SME fleet replacement, wallbox packages, ESG taxonomy checks, subsidy caveats, tax reimbursement, or small-business document requests, and the system answers in a predictable structure that leasing employees can review and reuse.
Every generated answer is grounded in approved knowledge-base sources and includes citations that can be inspected in the sources panel. The frontend normalizes answer sections into a compact table, highlights citations inline, and keeps the cited evidence available for follow-up questions so employees can trace how a recommendation was formed.
Hybrid Retrieval Pipeline
The backend combines document ingestion, chunking, source inventory metadata, lexical BM25 retrieval, and optional dense Faiss search. Topic profiles expand queries with domain-specific terms for operational leasing processes, charging infrastructure, EU Taxonomy, green claims, subsidies, reimbursement, tax, and GDPR-sensitive fleet data.
This retrieval layer helps the assistant distinguish between product/process guidance, regulatory criteria, and synthetic demo examples. It also preserves source title, URL, retrieval date, freshness status, section metadata, and review hints, giving employees a practical route from conversational output back to the underlying evidence.
Guardrailed Case Workspace
The user experience is designed as a case workspace rather than a generic chatbot. Recent cases are stored locally, the conversation panel streams the assistant response in real time, and the sources panel collects evidence used across answers. Suggested prompts guide employees toward high-value workflows such as EV fleet assessment, charging infrastructure documents, subsidy caveats, and client reply drafting.
Domain guardrails keep the product aligned with internal support boundaries. The assistant can prepare checklists, missing-fact questions, caveats, and draft communications, but it does not approve credit, provide final legal or tax advice, guarantee subsidies, or make final taxonomy-alignment claims. That restraint makes the tool more suitable for regulated leasing operations and internal review workflows.
Contact Us
Anastasia Timoshenko
Regional Account Manager
700+
In-house developers
27+
Years in industry
300+
Clients world wide
27+
Years in industry
5.0 Ratings