Python / LLM Expert: AI Document Analysis Review in 3-5 Days
Budget / Salary$30–250
TypeFreelance project
LocationRemote
Posted2 hours ago
I’m looking for an experienced independent AI engineer to review and improve an existing Python document-analysis and decision-support prototype. It processes PDFs containing text, tables, diagrams and measurements, retrieves relevant source material, and produces structured recommendations with supporting citations.
The core implementation is already in place:
- PDF text and image extraction, with a local searchable document index.
- Adaptive retrieval and source citations.
- Model integration, caching and bounded workflow steps.
- Output validation, safeguards and human-review fallback behavior.
- Detailed traces, replay tools, automated checks and live-model experiments.
My current internal readiness estimate is approximately 70/100, with a target of 90–95/100 supported by measurable improvements. These are readiness targets, not measured prediction-accuracy percentages or a guaranteed outcome.
The main areas needing expert investigation are:
1. Retrieval coverage, chunking and context selection, including exceptions and cross-references.
2. Reliable interpretation of tables, diagrams and numerical measurements.
3. Checking that cited evidence actually supports each conclusion and preserves qualifications or negation.
4. Distinguishing genuinely missing evidence from issues already resolved by the available documents, with appropriate human-review fallback.
5. End-to-end evaluation on fresh scenarios, while balancing accuracy, latency, cost and maintainability.
Expected deliverables within the next three days:
- A concise root-cause assessment and prioritized improvement plan.
- Agreed, reviewable code or prompt improvements with relevant tests.
- Reproducible before/after results, including regressions and remaining limitations.
- A short technical handover explaining the changes.
I prefer someone with a strong verified track record in Python, retrieval-augmented generation, multimodal document processing and LLM evaluation. Please include one or two relevant examples, your proposed approach, earliest availability and fixed-price quote for this scope.
Initial discussion and screening will use synthetic or redacted examples. Access to any additional materials will be considered separately, subject to permission and confidentiality requirements.
The core implementation is already in place:
- PDF text and image extraction, with a local searchable document index.
- Adaptive retrieval and source citations.
- Model integration, caching and bounded workflow steps.
- Output validation, safeguards and human-review fallback behavior.
- Detailed traces, replay tools, automated checks and live-model experiments.
My current internal readiness estimate is approximately 70/100, with a target of 90–95/100 supported by measurable improvements. These are readiness targets, not measured prediction-accuracy percentages or a guaranteed outcome.
The main areas needing expert investigation are:
1. Retrieval coverage, chunking and context selection, including exceptions and cross-references.
2. Reliable interpretation of tables, diagrams and numerical measurements.
3. Checking that cited evidence actually supports each conclusion and preserves qualifications or negation.
4. Distinguishing genuinely missing evidence from issues already resolved by the available documents, with appropriate human-review fallback.
5. End-to-end evaluation on fresh scenarios, while balancing accuracy, latency, cost and maintainability.
Expected deliverables within the next three days:
- A concise root-cause assessment and prioritized improvement plan.
- Agreed, reviewable code or prompt improvements with relevant tests.
- Reproducible before/after results, including regressions and remaining limitations.
- A short technical handover explaining the changes.
I prefer someone with a strong verified track record in Python, retrieval-augmented generation, multimodal document processing and LLM evaluation. Please include one or two relevant examples, your proposed approach, earliest availability and fixed-price quote for this scope.
Initial discussion and screening will use synthetic or redacted examples. Access to any additional materials will be considered separately, subject to permission and confidentiality requirements.
Apply on Freelancer →
Project sourced from Freelancer.com. Applications happen directly on the original platform — we never collect your data.