Browse categories
Explore
Fiverr Pro
English
$
USD
I will build a custom ai rag chatbot with langchain, openai, and source citations
Loc T
About this gig
Tired of generic AI chatbots that hallucinate fake facts and make up answers?
I build custom, enterprise-grade Retrieval-Augmented Generation (RAG) AI assistants powered by Next.js, Django REST Framework, and Cohere/OpenAI. Your chatbot strictly answers based on your private company documents, PDFs, manuals, or website data with verified citations.
Core Capabilities:
- 3-Stage Hybrid Retrieval: Dense Vector Cosine + Lexical BM25 fused via Reciprocal Rank Fusion (RRF)
- Neural Cross-Encoder Reranking: Cohere Rerank v3.5 ensures top-tier context precision
- 100% Verified Citations: Interactive [1], [2] pill badges with popover source previews
- Anti-Hallucination Guardrails: Tells users when documents lack info rather than confabulating
- Tree-Based Conversation Branching: Edit prompts & switch message versions
- Real-time SSE Token Streaming: Word-by-word instant chat response
- Full Document Management: Built-in web scraper & document uploader
- Production Ready: 1-click Docker Compose setup with Nginx reverse proxy
Tech Stack: Next.js, React, Tailwind CSS, Python, Django, Cohere/OpenAI, VectorDB, Docker.
Message me before ordering to discuss your documents and knowledge base formats!
Get to know Loc T
Loc T
Full Stack Developer
- FromVietnam
- Member sinceMay 2024
- Last delivery1 year
Languages
English, Vietnamese
All websites i made:
- RAG chatbot system (django, reactjs, vectordb, openai).
- DeepDoc - similar to gdrive (django, vuejs, elasticsearch, mysql, vncorenlp).
- DeepSpelling - similar to Grammarly but for Vietnamese language (django, vuejs, add-in word, mysql, transformer).
- DocSearch (django, vuejs, elasticsearch, vncorenlp-underthesea, ocr).
My Portfolio
FAQ
How do you prevent the AI from hallucinating?
The system utilizes strict prompt grounding and a 3-stage hybrid retrieval pipeline (Dense Vector + BM25 + Cross-Encoder Reranking). If the answer is not present in your provided documents, the bot is instructed to state that the context does not contain the answer rather than guessing.
What document formats can I upload into the knowledge base?
You can upload raw text, Markdown, PDF documents, or use the built-in web scraper to ingest URLs directly into vector embeddings.
Can I connect this to OpenAI (ChatGPT) or Anthropic Claude instead of Cohere?
Yes! While Cohere Command-R and Rerank v3.5 offer state-of-the-art grounded RAG, the architecture is modular and can easily be connected to OpenAI (GPT-4o) or Anthropic Claude.
Will I receive the complete code and deployment instructions?
Yes! You get full source code ownership (frontend and backend) with a 1-click ./run.sh setup script and a production-ready docker-compose.yml file for self-hosting on your VPS (DigitalOcean, AWS, Hetzner).
