Build AI Evidence Verification PoC

Python Software Engineer Required – AI Research and Evidence Verification Proof of Concept

Project Type: Fixed-price development
Location: Remote – Worldwide
Budget: Competitive quotations requested
Project Stage: Existing experimental codebase with a smaller proof of concept required

Project Overview

We are seeking an experienced Python software engineer to develop a working proof of concept for a local-first AI-assisted research and evidence-verification system.

The objective is to demonstrate that AI-generated technical research can be checked against independently retrieved and preserved source material, with unsupported factual claims identified and prevented from being presented as verified.

This is a focused software engineering project, not a general chatbot or conventional retrieval-augmented generation application.

A larger system architecture and experimental codebase already exist. We now wish to develop a substantially smaller, independently testable implementation that demonstrates the core concept before committing to full-scale development.

Required Proof of Concept

The initial system must demonstrate the following capabilities.

1. Research Input

Accept a technical question or research problem and preserve the original request.

2. Evidence Acquisition

Retrieve relevant publicly available source documents through an approved search API and independently save the original accessible material.

3. Evidence Preservation

Record source URLs, retrieval timestamps, original files and cryptographic hashes.

4. AI-Assisted Extraction

Use a locally executed language model to identify candidate factual statements from saved evidence.

5. Independent Verification

Use deterministic Python code to check extracted quotations and references against saved evidence, rejecting fabricated or mismatched citations.

6. Existing-Solution Investigation

Perform a limited search for previously documented technical solutions, products or mechanisms relevant to the research question.

7. Owner-Controlled Decisions

Present potential equivalent solutions for human review. The software must never automatically discard a proposed direction based solely on an AI assessment.

8. Evidence-Linked Reporting

Generate a readable report identifying supported claims, source references, conflicting evidence, failed searches and unresolved questions.

The proof of concept must distinguish verified source content from AI-generated reasoning and untested assumptions.

Preferred Technology

Python, LM Studio or compatible local LLM inference, REST API integration, pytest, SHA-256 file integrity verification, local file storage, Git, GitHub and Windows 11 compatibility.

We prefer existing, well-maintained libraries where appropriate rather than unnecessary custom development.

Developers may recommend alternative technical approaches where they can demonstrate improved reliability, reduced cost or simpler implementation.

Development Requirements

The project will be divided into small, independently verifiable milestones.

Each milestone must include working source code, documented installation instructions, automated positive and negative tests, reproducible test results, identified limitations and a demonstration of the completed functionality.

AI-assisted development tools may be used. However, developer explanations or AI-generated assurances will not be accepted as evidence that software functions correctly.

Acceptance will depend on executable evidence and independently reproducible testing.

The completed proof of concept must operate on a standard Windows laptop using an approved local language model.

Confidentiality and Intellectual Property

This project involves confidential research and proprietary software concepts.

The selected developer must agree to appropriate confidentiality and intellectual-property assignment terms before receiving non-public technical materials.

The initial assignment will use a limited and isolated development scope. Access to the complete architecture, original private repository and protected acceptance tests will not be provided.

Ownership of commissioned source code and agreed deliverables must be transferred to the commissioning party under the contract.

Subcontracting or disclosure of confidential information will require prior written approval.

Information Required From Applicants

Please address the following questions in your proposal.

1. Have you developed Python applications involving local language models, deterministic validation, source verification or evidence-based research? Please provide relevant examples.

2. How would you structure a small, end-to-end proof of concept of this nature?

3. Which existing Python libraries or frameworks would you recommend using?

4. How would you independently verify that fabricated quotations, incorrect citations and unsupported factual assertions are rejected?

5. What fixed price would you propose, including an itemised milestone breakdown?

6. What development timeframe would you estimate, and which technical assumptions affect that estimate?

7. Would you be willing to complete a small paid technical evaluation before being awarded the main project?

Selection Criteria

We will prioritise demonstrated engineering experience, systematic debugging, secure implementation, independent testing, practical problem-solving and clear communication.

We are seeking the most cost-effective engineer capable of delivering independently verifiable results, rather than simply the lowest-priced applicant.

Please begin your proposal with the words Evidence Verification PoC so we can distinguish relevant applications from generic submissions.

---
來源：Freelancer.com | 出價次數：0 | 剩餘時間：6 日 | 專案類型：固定預算