GS
AI Research Intern Agricultural Knowledge Systems
gsarp · Lalitpur · onsite · Posted 52d ago
Not disclosedFresherfulltimeSourced from
About a minute. You'll make an account on the way through.
About this role
About The Role
Role Overview
Technical Requirements
Role Overview
- Location: Pulchowk, Lalitpur, Nepal (On-site/Hybrid)
- Duration: 3 Months
- Type: Internship (with Research or Junior Engineer Progression)
- Focus: Building a high-fidelity Neurosymbolic knowledge base from thousands of government-validated agricultural PDFs to support decision-making for international stakeholders (FAO, WFP, IFAD).
- Advanced Document Parsing: Implement extraction pipelines using Docling and PyTesseract to convert complex, scanned Nepali government documents into structured Markdown while preserving tables and hierarchical headings.
- Knowledge Engineering: Design and implement algorithms to explicitly extract Subject-Predicate-Object (SPO) triples (e.g., Jumli Marshi -> hasAttribute -> Cold Tolerance) from unstructured Devanagari text.
- Linguistic Processing: Utilize NLP libraries such as Stanza or spaCy for Nepali dependency parsing to identify grammatical relationships and improve the accuracy of triple extraction in low-resource settings.
- Semantic Mapping & Linking: Map extracted agricultural entities to global standards like AGROVOC or Wikidata to ensure interoperability and resolve synonyms across different ministry reports.
- GraphRAG Implementation: Assist in developing a GraphRAG (Graph-based Retrieval-Augmented Generation) architecture that enables multi-hop reasoning over the knowledge graph to answer complex agricultural queries.
- Data Validation & Oversight: Participate in human-in-the-loop validation sessions to refine the knowledge graph's accuracy, ensuring the system meets the high verification standards of national agricultural ministries.
Technical Requirements
- AI Stack: Strong Python skills with experience in NLP frameworks (Transformers, spaCy, or NLTK).
- OCR & Extraction: Hands-on experience with Tesseract and digital document parsers; familiarity with handling Devanagari script or non-Unicode fonts (e.g., Preeti) is a plus.
- Graph Technologies: Basic understanding of Graph Databases (**Neo4j **or similar) and semantic triples (RDF).
- Domain Interest: Passion for applying AI to agricultural development, food security, and regional challenges in Nepal.
Sourced from foundit
Skills
digital document parserssemantic triplesNltkSpacyNeo4Jgraph databasesHugging Face TransformersTesseractNLP frameworksRDFPython