Contact person: AI/ML + GIS Developer for CoT-based LLM System

Phone:Show

Email:Show

Location: Pune, India

Budget: Recommended by industry experts

Time to start: As soon as possible

Project description:
"We are looking for an experienced AI/ML + GIS developer to design and implement a Chain-of-Thought (CoT)-based LLM system that can solve complex spatial analysis tasks through intelligent geoprocessing orchestration.

Context:
Geospatial problems such as flood risk assessment, site suitability analysis, and land cover monitoring often require expert-designed workflows across multiple GIS tools and datasets. Our aim is to build a system where LLMs act like a geospatial expert, automatically planning and executing workflows step by step from natural language queries.

Project Objectives:

Develop a reasoning-enabled framework that combines LLMs and geoprocessing APIs to auto-generate multi-step GIS workflows.

Support integration of heterogeneous spatial datasets (OpenStreetMap, Bhoonidhi, etc.) and libraries for dynamic tool/resource selection.

Build an interface that converts user queries into sequential GIS tasks, while transparently logging Chain-of-Thought reasoning.

Demonstrate with benchmark tasks like flood mapping and site suitability analysis, including visualization of results.

Expected Deliverables:

Web/Desktop application.

Outputs:

JSON/YAML workflow definitions.

Chain-of-Thought reasoning logs.

GIS outputs & visualizations.

Demonstrations on flood risk mapping and site suitability.

Metrics comparing against manual baselines (accuracy, runtime, errors, resource usage).

Tools/Technologies (Suggested):

LLMs: Mistral-7B, LLaMA-3-8B, Phi-2 (integrated via LangChain/Transformers).

GIS Tools: QGIS, GRASS GIS, GDAL, Rasterio, GeoPandas, WhiteboxTools, PyQGIS.

Frameworks/UI: Retrieval-Augmented Generation (RAG), JSON schema for GIS functions, Streamlit for UI/visualization.

Key Features to Implement:

Prompt library for Chain-of-Thought reasoning.

RAG with GIS documentation (QGIS, GRASS, GDAL).

JSON schema defining GIS functions with parameter selection.

Reasoning–Acting–Observation loop for iterative improvement.

Error handling for CRS mismatches, geometry issues, etc.

Evaluation Parameters:

Logical + syntactic validity of workflows.

Clarity & traceability of Chain-of-Thought reasoning.

Robustness across a wide range of queries.

Efficiency, usability, and accuracy compared to manual workflows.

Ideal Freelancer:

Strong background in AI/ML (LLMs, LangChain, RAG, reasoning systems).

Hands-on experience with GIS tools & APIs (QGIS, GDAL, GeoPandas, Rasterio).

Skilled in Python development .

Experience with data integration and workflow automation." (client-provided description)