Agentic RAG Platform & Databases
Budget: $5 – $70 NZD
I need an end-to-end Retrieval-Augmented Generation platform, powered by Ollama-hosted language models, that lets a user drop in a document—TXT, PDF, or DOCX—then automatically searches my custom knowledge bases for exact matches and contextually related material, finally returning a concise, well-structured summary.
Scope
• Data ingestion: raw folders of text must be cleaned, chunked, embedded and stored in a vector store you recommend (Faiss, Milvus, or similar).
• User workflow: drag-and-drop upload, progress feedback, and a chat-style window that displays the sourced passages beside the generated synopsis.
• Retrieval logic: combine exact-match look-ups with semantic search so the AI can cite both literal hits and topic-level associations.
• Agentic orchestration: the system should chain tasks—retrieval, ranking, summarisation, and reference insertion—without manual prompts.
• Deployment: containerised (Docker) so I can spin it up on my own GPU server alongside Ollama.
Acceptance criteria
1. A running local instance that processes at least 1 GB of mixed documents in under five minutes.
2. Summaries must display clickable citations that trace back to the source paragraph.
3. A README explaining setup, environment variables, and how to add new data sets.
If you have shipped similar RAG stacks before and can show a quick demo link or repo, that will move things along quickly.
Scope
• Data ingestion: raw folders of text must be cleaned, chunked, embedded and stored in a vector store you recommend (Faiss, Milvus, or similar).
• User workflow: drag-and-drop upload, progress feedback, and a chat-style window that displays the sourced passages beside the generated synopsis.
• Retrieval logic: combine exact-match look-ups with semantic search so the AI can cite both literal hits and topic-level associations.
• Agentic orchestration: the system should chain tasks—retrieval, ranking, summarisation, and reference insertion—without manual prompts.
• Deployment: containerised (Docker) so I can spin it up on my own GPU server alongside Ollama.
Acceptance criteria
1. A running local instance that processes at least 1 GB of mixed documents in under five minutes.
2. Summaries must display clickable citations that trace back to the source paragraph.
3. A README explaining setup, environment variables, and how to add new data sets.
If you have shipped similar RAG stacks before and can show a quick demo link or repo, that will move things along quickly.