Part-Time Policy Lens Consultant

The consultant will design and pilot a small set of policy lenses that extract structured, comparable representations from policy documents, and demonstrate their value on a bounded corpus from a small number of ESCWA member States, in a form that ESCWA can evaluate and extend afterward. The assignment is a feasibility pilot, not a production system: one policy domain, three to five countries, three to five lenses, evaluated qualitatively with policy-expert input.

Remote

6 months

Background: The Decision Support and Data Science Division (DSDS) of the United Nations Economic and Social Commission for Western Asia (ESCWA) seeks a part-time consultant to design and pilot Policy Lens, a structured evidence-synthesis approach for policy documents. Comparing how several countries approach a policy issue means reading many national strategies, evaluations and progress reviews, written years apart and with uneven rigour. Patterns across the set: shared causal assumptions, claims backed by evidence versus simply asserted, targets - are hard to see even when the information is in the text. Policy Lens addresses this by applying an analytical lens to the whole document set at once. Each lens rests on a frame: a defined structure with specific fields that every document maps onto, so outputs are comparable across documents rather than readable one at a time. Examples include a Theory of Change lens, an evidence-strength lens and a gap-detection lens. ESCWA already operates two LLM-based systems: Mustashar, a retrieval-augmented question-answering system, and Naseej, an interactive multi-agent system that drafts substantive policy reports and briefs based on a UN knowledge corpus. Both produce prose for a human reader; neither produces structured, comparable representations. Policy Lens is intended as complementary infrastructure that can feed more reliable, structured material into both. The intended source corpus is the Arab Development Portal (ADP), whose Knowledge Hub documents are already tagged by country, SDG and theme, alongside structured indicator data and an SDG data-availability tracker. This lowers the extraction burden for descriptive and gap-detection lenses in a pilot. Duties and responsibilities: Under the supervision of the Chief of DSDS, the consultant will: 1. Design the lens framework. Articulate the lens/frame concept and develop a taxonomy of lenses (descriptive/comparative, causal, gap-detection), with a justification for Theory of Change as the starting point for the causal category. 2. Apply political-neutrality safeguards. Build neutrality into lens design, especially gap-detection lenses, framing findings as opportunities for regional learning or support rather than as country scorecards. 3. Scope the pilot corpus with ESCWA. Agree on one policy domain and three to five countries and select well-structured English-language documents (national strategies, official reports) from ADP and other sources ESCWA provides. 4. Specify and test three to five pilot lenses. Include at least one descriptive lens (e.g. SDG Alignment), one causal lens (e.g. Theory of Change) and one gap-detection lens (e.g. Population Coverage), each with a defined extraction schema. 5. Build a simple extraction and comparison pipeline. Implement three stages — extraction, structuring into the lens schema, output generation — using LLM APIs within the budget ESCWA confirms, and leveraging existing ADP metadata and indicator data where possible. 6. Produce structured outputs for policymakers. Generate tables, comparison views and simple visualizations (e.g. a theory-of-change diagram per country). 7. Run an expert validation round. Coordinate a lightweight qualitative review of outputs by experts ESCWA identifies, focused on substantive accuracy, and document findings and error patterns. 8. Recommend integration paths. Assess concretely how Policy Lens outputs could feed Naseej (as structured source material for briefs) and Mustashar (as a structured index alongside text retrieval), in consultation with the teams that own those systems. 9. Document and hand over. Deliver code, schemas, prompts and documentation in ESCWA's repositories so the work can be reproduced and extended. 10. Prepare a methods paper (best effort). Draft a methods-oriented paper on the lens/frame approach, grounded in pilot results, for possible submission with ESCWA clearance.

A Master's degree or equivalent in computer science, data science, computational linguistics, public policy, economics or related area is required. A PhD is desirable. All candidates must submit a copy of the required educational degree. Incomplete applications will not be reviewed. At least five years of progressively responsible experience in applied NLP, LLM-based systems or data science is required. Demonstrated experience building LLM pipelines for information extraction or structured output (schema-constrained generation, prompt design, evaluation) is required. Proficiency in Python and common LLM/NLP tooling is required. Experience analyzing public policy documents, evidence synthesis, or Theory of Change and results frameworks is desirable. Familiarity with the SDG framework and development indicators is desirable. Experience with retrieval-augmented or multi-agent systems is desirable. Publications in NLP, computational social science or policy analytics is desirable. Prior work with the UN system or other international organizations, particularly in the Arab region is desirable.

English and French are the working languages of the United Nations Secretariat; and Arabic is a working language of ESCWA. For this position, fluency in English is required. Note: “Fluency” equals a rating of ‘fluent’ in all four areas (speak, read, write, and understand) and “Knowledge of” equals a rating of ‘confident’ in two of the four areas.

Not available.

THE UNITED NATIONS DOES NOT CHARGE A FEE AT ANY STAGE OF THE RECRUITMENT PROCESS (APPLICATION, INTERVIEW MEETING, PROCESSING, OR TRAINING). THE UNITED NATIONS DOES NOT CONCERN ITSELF WITH INFORMATION ON APPLICANTS’ BANK ACCOUNTS.