Internship Task - AgenticAI

The document outlines a task-oriented AI internship project focused on building an agentic AI system using LangChain and a foundational language model. The project involves setting up the environment, developing a functional agent with integrated tools, creating a user interaction interface, and evaluating the agent's performance on real-world queries. Deliverables include a Python script or Jupyter notebook, deployment instructions, a demo video, and a README file detailing the setup and tool choices.

Uploaded by

mananarshad351

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

6 views3 pages

Internship Task - AgenticAI

Uploaded by

mananarshad351

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

You are on page 1/ 3

KDD LAB Agentic AI Internship Task

Project Title:

Task-Oriented AI Agent Using LangChain and Tool Integration

Objective:

Build a functional agentic AI system using LangChain and a foundational language model
(LLM) that can solve user-defined tasks by utilizing memory, tools (e.g., search, calculator),
and reasoning. This task will help you explore agent design, multi-step reasoning, and
decision chaining, which are core to agentic AI systems like personal assistants, coding
agents, and planning bots.

Task Breakdown

Part 1: Environment Setup and Tool Familiarization

● Install and set up:

○ Python (3.9+)
○ Core libraries: langchain, openai or ollama, faiss, pandas, requests
○ Optional tools: ChromaDB, DuckDuckGoSearch, PythonREPLTool,
SerpAPI, etc.

● Explore LangChain’s agent types: Zero-shot, ReAct, and Conversational agents.

● Create a basic agent that can answer simple questions using OpenAI or a local LLM
via Ollama or Hugging Face Transformers.
Part 2: Build a Functional Agent with Tools

● Define a task-oriented agent (e.g., research assistant, report summarizer, personal

finance advisor).
● Integrate at least 3 tools with the agent:
○ A search tool (e.g., DuckDuckGoSearch or SerpAPI)
○ A Python REPL tool
○ A memory component (e.g., conversational memory or vector store)

● Define a prompt template for guiding the agent’s reasoning.

● Allow multi-step reasoning — the agent should autonomously decide which tools to
use and in what order.

Part 3: User Interaction Interface (Optional Task)

● CLI or Streamlit/Flask interface:

○ Allow users to input a query (e.g., "Summarize latest news on AI regulation
and give me a cost estimate of deploying a model").
○ Show each reasoning step and the final result.
● Display:
○ Intermediate decisions
○ Final outcome
○ Tool usage log

Part 4: Evaluation & Analysis

● Test your agent on 5+ real-world queries or use cases (e.g., trip planning, investing
assistant, resume enhancements).

● Evaluate:
○ Tool selection efficiency
○ Relevance of outputs
○ Error handling when the agent fails or gets stuck

● Analyze:

○ Agent’s behavior across prompt variations

○ Memory effectiveness
○ Limitations of current LangChain-based systems

Bonus Part (Completely Optional)

● Add long-term memory with a vector database (e.g., Chroma or FAISS).

● Use LangGraph to model and visualize the agent’s workflow graphically.
● Extend your agent to trigger APIs or execute commands (with safeguards).

Deliverables

● Python script or Jupyter notebook implementing the agent.

● If a web UI is built, provide deployment instructions (localhost or GitHub).
● Short video walkthrough (Demo) of your system:

○ Agent architecture & components

○ Tool integration
○ Query examples and live demonstration

● Include a README file with:

○ Setup steps
○ Tool choices and rationale
○ Known issues (if any arise) and next steps
● GitHub repo encouraged to be submitted alongside.

Disclaimer & Development Integrity Guidelines

This task is meant to test your core understanding of agent-based system design and the
LangChain ecosystem. Please adhere to the following:

● Do not use GPT, Claude, Gemini, Copilot, or similar tools to generate the full code
or entire architecture without core understanding.
● You may consult tutorials or documentation, but the final code and system logic
must be your own.
● You will be required to explain each part of your implementation, especially the
reasoning engine, prompt design, and tool integrations.
● Submissions that include unexplained, AI-generated, or copied code will be
disqualified.

Key Libraries & Frameworks You May Use

● Python – Core development language

● LangChain – Agent framework for tool use, chaining, memory
● PyTorch / TensorFlow – (Optional) if building or fine-tuning models
● scikit-learn / pandas / NumPy – Data processing
● FAISS / Chroma – Vector stores for memory
● Streamlit / Flask – UI development