Local AI Agent for Research devolopment

In need for a skilled AI developer to create a an AI Agent for research, its core goal is to be an assistant system for me. This system will process and generate responses solely from documents that I upload. The use of DeepSeek or Chat GPT 12x7B or a comparable high-accuracy model is required. Key features should include:
you have the freedom to use any of the following or suggest an alternative to establish my requirements :

Option 1: Custom Chat UI + Retrieval
Option 2: Pre-built RAG System + DeepSeek 12x7B
Option 3: Quivr AI + DeepSeek 12x7B


- Long-context processing
- Retrieval-augmented generation (RAG)
- Workspace-based document organization
- Chain-of-thought reasoning
- Citation generation

The ideal system would be cost effective, scalable, and easy to maintain. I am open to various implementation strategies and have pinpointed three potential solutions. I am seeking a developer or team who can not only recommend the optimal path but also carry out the development.

Please note that I prioritize cost-effectiveness above all. The types of documents the system will handle are varied, and include PDF, Word, text, Jpeg, Png,

Key Features & Requirements

Use Only Uploaded Documents – No external data sources should influence responses.
(Chat GPT or Deepseek or any other) 12x7B or Similar High-Accuracy Model – Ensures logical coherence, reduced hallucination, and precision in academic content generation.
Multiple Workspaces ("Spaces") – Users can create separate document groups for different research topics or projects.
Workspace Selection – Users must be able to choose which space(s) the system should use for responses.
Stateful Conversation Memory Per Space – The system retains context across multiple interactions within the same space.
Retrieval-Augmented Generation (RAG) – Responses must be based on retrieved excerpts from the uploaded documents.
Chain-of-Thought Prompting – Model must break down complex queries into structured, step-by-step logical reasoning.
Interactive Feedback Loops – Users should be able to review and approve model-generated reasoning before finalizing the response.
Long-Context Handling (10,000+ Tokens) – The system must support long academic documents (e.g., 300-pages ).
Automatic Citation Generation – Extract references from uploaded documents and embed citations in responses.
High Accuracy and Logical Coherence – Ensures responses are detailed, structured, and aligned with scholarly research.
Fine-tuning for Domain-Specific Accuracy – The model should be optimized for public health, behavioral studies, psychology, social scince and other specialized research areas.


Note : before you start you and I must layout the protochol for each step this AI agent must follow,