harshith-7/discover-agent

1# 🔎 Discover Agent
2
3Discover Agent is an AI-powered web search application that utilizes **LangGraph Agents**, **DuckDuckGo** search, and Groq-hosted **DeepSeek-R1 Distill LLaMA 70B** to dynamically fetch and synthesize information. Unlike traditional workflows, this project dynamically decides when, how and whether to use tools during execution, showcasing the power of autonomous **AI agents** in navigating uncertain tasks.
4
5![img.png](img.png)
6
7---
8
9## 🚀 Project Overview
10
11This project creates a **Discover Agent** that:
12
13* Accepts natural language queries via a web UI,
14* Uses DeepSeek-R1 LLaMA 70B Distill (hosted on Groq) as the reasoning LLM,
15* Employs tools such as:
16  * DuckDuckGo Search for real-time web lookup, 
17  * Datetime tool to fetch current date/time if needed,
18* Routes the conversation intelligently via LangGraph,
19* Displays real-time, concise answers.
20
21## ✨ Features
22
23* 🌐 Real-time Web Search (DuckDuckGo)
24* 🕓 Current Date & Time Tool
25* 🧭 deepseek-r1-distill-llama-70b via Groq for planning and response generation
26* 🧱 LangGraph Agent with tool invocation and conditional logic
27* 🔄 Stateful and flexible architecture
28* 🖥️ Streamlit-based user interface
29* ⚡ FastAPI backend
30
31---
32
33## 🧠 What is an AI Agent?
34
35An **AI Agent** is a self-directed system that:
36
37* Interprets complex input (e.g., human questions),
38* Perceives its environment (e.g., user queries and tools),
39* Decides what actions (e.g., tools to use) to take independently,
40* Acts autonomously (e.g., fetches and processes information),
41* And adapts based on feedback or reasoning loops.
42
43LangGraph provides a framework to build these agents with dynamic, branching logic and persistent memory, surpassing traditional workflows. LangGraph agents go beyond static workflows by supporting conditional branching, autonomous tool use decisions, and persistent state handling.
44
45## 🔄 LangGraph vs. LangChain: Why LangGraph?
46
47| Feature             | LangGraph                                | LangChain (Core)                    |
48| ------------------- | ---------------------------------------- | ----------------------------------- |
49| Execution Model     | **Graph-based execution**                | Linear or chain-of-tools            |
50| Tool Decision Logic | Built-in conditional routing (agents)    | Requires manual logic or chains     |
51| State Persistence   | **Explicit state tracking**              | Implicit through variables          |
52| Flexibility         | Better for **multi-step, dynamic logic** | Suitable for **simple tool chains** |
53
54**LangGraph** is ideal when you need an agent that decides *when* to use a tool and *how* to proceed after tool output—like in this project.
55
56## 🧠 Discover Agent vs. Discover (Workflow)
57
58Here is a [link](https://github.com/harshith-77/discover) to the previous version "Discover" for your reference
59
60| Feature             | Discover Agent (LangGraph)                | Discover Workflow (Earlier Version)    |
61| ------------------- | ----------------------------------------- | -------------------------------------- |
62| Tool Decision       | Dynamic & Conditional                     | Predefined Steps                       |
63| Extensibility       | Easy to plug in new tools / logic         | Hardcoded path                         |
64| Autonomy            | Agent decides next step                   | Linear, no adaptive logic              |
65| Complexity Handling | Better for multi-turn or tool-heavy tasks | Suitable for simple single-step tasks  |
66| Maintenance         | Modular and easier to extend              | Less flexible, more code change needed |
67
68**Drawbacks of agent-based approach:**
69
70* Slightly more overhead to set up initially
71* Graph logic can be overkill for very simple tasks
72
73---
74
75## ⚙️ Installation
76
77### 1. Clone the repository
78
79```bash
80git clone https://github.com/harshith-77/discover-agent.git
81cd discover-agent
82```
83
84### 2. Install dependencies
85
86```bash
87pip install -r requirements.txt
88```
89
90### 3. Set up environment variables
91
92Create a `.env` file with:
93
94```
95GEMINI_API_KEY=your_google_genai_key
96```
97
98## ▶️ Run the App
99
100### Start the FastAPI backend:
101
102```bash
103python main.py
104```
105
106### Launch the Streamlit UI:
107
108```bash
109streamlit run main_ui.py
110```
111
112## 🧰 Tech Stack
113
114| Layer       | Technology                                 |
115| ----------- |--------------------------------------------|
116| LLM         | DeepSeek-R1 LLaMA 70B Distill via Groq API |
117| Agent       | LangGraph                                  |
118| Tool        | DuckDuckGoSearchRun, Custom Datetime Tool  |
119| UI          | Streamlit                                  |
120| Backend     | FastAPI + Uvicorn                          |
121| Env Manager | python-dotenv                              |
122
123---
124
125## 📁 File Structure
126
127```
128├── main.py           # FastAPI backend
129├── main_ui.py        # Streamlit UI
130├── helper.py         # Agent logic with LangGraph
131├── .env              # API keys (not checked into git)
132├── requirements.txt  # Python dependencies
133```
134
135## 🗣 Example Usage
136
1371. Open the Streamlit app.
1382. Type a query like:
139   **"Who won the IPL 2025 final?"**
1403. The agent provides a concise response to the query by fetching from web if needed.
141
142---
143
144## 📜 License
145
146MIT License
147