Skip to content

A curated collection of practical AI projects implementing OCR systems, RAG, AI agents, and other AI use cases.

feat: add Multi-Agent Coding Assistant using Mistral and LangChain

Tiioluwanid83ea4da7dd8
diff --git a/ai_agents/multi_agent_coding_assistant/.env.example b/ai_agents/multi_agent_coding_assistant/.env.example
new file mode 100644
index 0000000..08db593
--- /dev/null
+++ b/ai_agents/multi_agent_coding_assistant/.env.example
@@ -0,0 +1,2 @@
+# Mistral AI API key — obtain one at https://platform.mistral.ai
+MISTRAL_API_KEY=your_mistral_api_key_here
diff --git a/ai_agents/multi_agent_coding_assistant/README.md b/ai_agents/multi_agent_coding_assistant/README.md
new file mode 100644
index 0000000..8cffb99
--- /dev/null
+++ b/ai_agents/multi_agent_coding_assistant/README.md
@@ -0,0 +1,107 @@
+# Multi-Agent Coding Assistant
+
+> A Streamlit app where three specialized AI agents collaborate to plan, write, review, and refine Python code from a single natural-language task description.
+
+## Overview
+
+Multi-Agent Coding Assistant orchestrates a four-stage pipeline powered by Mistral AI. You describe what you want built; a Planner Agent structures the approach, a Coder Agent writes the first draft, a Reviewer Agent critiques it for bugs and improvements, and the Coder Agent produces a polished final version. All four stages are surfaced in an expandable Streamlit UI so you can inspect every step of the process.
+
+## Demo
+
+![Demo](assets/demo.png)
+
+**Example task:** *"Build a Python function that reads a CSV file and returns a dictionary of summary statistics (mean, median, std dev, min, max) for each numeric column."*
+
+**What you get:**
+- A structured implementation plan with function signatures and edge cases
+- A typed, documented first-draft implementation
+- A line-by-line code review covering bugs, performance, and best practices
+- A corrected final version with all reviewer feedback applied
+
+## Features
+
+- **Planner Agent** — breaks any coding task into a numbered, step-by-step implementation plan
+- **Coder Agent (Draft)** — translates the plan into typed, documented Python code
+- **Reviewer Agent** — audits the draft for bugs, edge cases, PEP 8 compliance, performance, and security issues
+- **Coder Agent (Final)** — applies every piece of reviewer feedback to produce production-ready code
+- Expandable result sections so you can compare draft vs. final side by side
+- Inline API key input as a fallback when no `.env` file is present
+
+## Tech Stack
+
+| Layer | Technology |
+|---|---|
+| LLM | Mistral Small 4 (`mistral-small-latest`) via Mistral AI API |
+| Agent Framework | LangChain (`langchain-mistralai`, `langchain-core`) |
+| UI | Streamlit |
+| Environment | python-dotenv |
+
+## Prerequisites
+
+- Python 3.9 or higher
+- A [Mistral AI API key](https://platform.mistral.ai)
+
+## Installation
+
+**1. Clone the repository**
+
+```bash
+git clone https://github.com/Sumanth077/Hands-On-AI-Engineering.git
+cd Hands-On-AI-Engineering/ai_agents/multi_agent_coding_assistant
+```
+
+**2. Create and activate a virtual environment**
+
+*Windows*
+```bash
+python -m venv .venv
+.venv\Scripts\activate
+```
+
+*macOS / Linux*
+```bash
+python -m venv .venv
+source .venv/bin/activate
+```
+
+**3. Install dependencies**
+
+```bash
+pip install -r requirements.txt
+```
+
+**4. Configure environment variables**
+
+```bash
+cp .env.example .env
+```
+
+Open `.env` and add your Mistral API key (see [Environment Variables](#environment-variables)).
+
+## Usage
+
+```bash
+streamlit run app.py
+```
+
+Open `http://localhost:8501` in your browser, describe a coding task, and click **Run Pipeline**.
+
+## Environment Variables
+
+| Variable | Description |
+|---|---|
+| `MISTRAL_API_KEY` | Your Mistral AI API key — obtain one at [platform.mistral.ai](https://platform.mistral.ai) |
+
+If `MISTRAL_API_KEY` is not set, the app will prompt you to paste it directly in the UI.
+
+## Project Structure
+
+```text
+multi-agent-coding-assistant/
+├── app.py # Streamlit UI and pipeline orchestration
+├── agents.py # LLM setup and agent prompt/invocation logic
+├── requirements.txt # Python dependencies
+├── .env.example # Environment variable template
+└── assets/
+ └── demo.png # App screenshot
+```
diff --git a/ai_agents/multi_agent_coding_assistant/agents.py b/ai_agents/multi_agent_coding_assistant/agents.py
new file mode 100644
index 0000000..f3685df
--- /dev/null
+++ b/ai_agents/multi_agent_coding_assistant/agents.py
@@ -0,0 +1,122 @@
+from langchain_mistralai import ChatMistralAI
+from langchain_core.messages import HumanMessage, SystemMessage
+
+
+PLANNER_SYSTEM_PROMPT = """You are a senior software architect and technical planner.
+
+Your job is to take a natural language coding task and break it down into a clear, structured implementation plan.
+
+Your plan must include:
+1. A brief summary of what will be built and why
+2. A numbered list of implementation steps in logical order
+3. For each step: what it does, what inputs it expects, and what it outputs
+4. Any edge cases or constraints to keep in mind
+5. Suggested function/class names and their responsibilities
+
+Be specific and actionable. The plan will be handed to a coder who will implement it exactly as described.
+Do not write any code — only the plan."""
+
+
+CODER_SYSTEM_PROMPT = """You are an expert Python developer who writes clean, production-ready code.
+
+Your job is to take an implementation plan and produce working Python code that fulfills it exactly.
+
+Requirements for all code you write:
+- Follow the plan step by step without skipping anything
+- Write clear, descriptive docstrings for every function and class
+- Add inline comments only where the logic is non-obvious
+- Handle all edge cases mentioned in the plan
+- Use type hints on all function signatures
+- Structure the code logically with related functions grouped together
+- Include example usage at the bottom inside an `if __name__ == "__main__":` block
+
+Return ONLY the Python code. No explanations before or after — just the code."""
+
+
+REVIEWER_SYSTEM_PROMPT = """You are a meticulous senior code reviewer focused on quality, correctness, and best practices.
+
+Your job is to review Python code and provide precise, actionable feedback.
+
+Your review must cover:
+1. **Bugs & Correctness** — logic errors, off-by-one errors, incorrect assumptions
+2. **Edge Cases** — missing input validation, unhandled exceptions, empty/None inputs
+3. **Best Practices** — PEP 8 compliance, naming conventions, code smells
+4. **Performance** — inefficient algorithms, unnecessary operations, memory issues
+5. **Security** — unsafe operations, injection risks, insecure defaults
+6. **Improvements** — cleaner implementations, missing features, better stdlib usage
+
+For each issue found, specify:
+- The exact location (function name or line description)
+- What the problem is
+- How to fix it
+
+Be direct and specific. The coder will use your feedback to rewrite the code."""
+
+
+FINAL_CODER_SYSTEM_PROMPT = """You are an expert Python developer who writes clean, production-ready code.
+
+Your job is to take an original implementation plan, a first draft of code, and a reviewer's feedback, then produce an improved final version of the code.
+
+Apply every piece of reviewer feedback. For each issue raised:
+- Fix bugs and logic errors exactly as suggested
+- Add missing edge case handling
+- Refactor for clarity and best practices
+- Improve performance where noted
+
+Requirements for the final code:
+- All reviewer concerns must be addressed
+- Maintain clear docstrings and type hints
+- Keep the `if __name__ == "__main__":` usage example, updated if needed
+- The final code must be complete and runnable
+
+Return ONLY the improved Python code. No explanations — just the final code."""
+
+
+def create_llm(api_key: str) -> ChatMistralAI:
+ return ChatMistralAI(
+ model="mistral-small-latest",
+ api_key=api_key,
+ temperature=0.1,
+ )
+
+
+def run_planner(llm: ChatMistralAI, task: str) -> str:
+ messages = [
+ SystemMessage(content=PLANNER_SYSTEM_PROMPT),
+ HumanMessage(content=f"Coding task: {task}"),
+ ]
+ response = llm.invoke(messages)
+ return response.content
+
+
+def run_coder(llm: ChatMistralAI, plan: str) -> str:
+ messages = [
+ SystemMessage(content=CODER_SYSTEM_PROMPT),
+ HumanMessage(content=f"Implementation plan:\n\n{plan}"),
+ ]
+ response = llm.invoke(messages)
+ return response.content
+
+
+def run_reviewer(llm: ChatMistralAI, code: str) -> str:
+ messages = [
+ SystemMessage(content=REVIEWER_SYSTEM_PROMPT),
+ HumanMessage(content=f"Review this Python code:\n\n```python\n{code}\n```"),
+ ]
+ response = llm.invoke(messages)
+ return response.content
+
+
+def run_final_coder(llm: ChatMistralAI, plan: str, draft_code: str, review: str) -> str:
+ messages = [
+ SystemMessage(content=FINAL_CODER_SYSTEM_PROMPT),
+ HumanMessage(
+ content=(
+ f"Original plan:\n\n{plan}\n\n"
+ f"First draft code:\n\n```python\n{draft_code}\n```\n\n"
+ f"Reviewer feedback:\n\n{review}"
+ )
+ ),
+ ]
+ response = llm.invoke(messages)
+ return response.content
diff --git a/ai_agents/multi_agent_coding_assistant/app.py b/ai_agents/multi_agent_coding_assistant/app.py
new file mode 100644
index 0000000..94b689c
--- /dev/null
+++ b/ai_agents/multi_agent_coding_assistant/app.py
@@ -0,0 +1,177 @@
+import os
+import streamlit as st
+from dotenv import load_dotenv
+from agents import create_llm, run_planner, run_coder, run_reviewer, run_final_coder
+
+load_dotenv()
+
+st.set_page_config(
+ page_title="Multi-Agent Coding Assistant",
+ page_icon="🤖",
+ layout="wide",
+)
+
+# ── Styles ────────────────────────────────────────────────────────────────────
+st.markdown(
+ """
+ <style>
+ .agent-badge {
+ display: inline-block;
+ padding: 2px 10px;
+ border-radius: 12px;
+ font-size: 0.78rem;
+ font-weight: 600;
+ margin-bottom: 6px;
+ }
+ .badge-planner { background:#dbeafe; color:#1e40af; }
+ .badge-coder { background:#dcfce7; color:#166534; }
+ .badge-reviewer { background:#fef9c3; color:#854d0e; }
+ .badge-final { background:#f3e8ff; color:#6b21a8; }
+ .stExpander > div:first-child { font-weight: 600; }
+ </style>
+ """,
+ unsafe_allow_html=True,
+)
+
+
+def _strip_code_fences(text: str) -> str:
+ """Remove markdown ```python ... ``` fences that the model may wrap around code."""
+ lines = text.strip().splitlines()
+ if lines and lines[0].startswith("```"):
+ lines = lines[1:]
+ if lines and lines[-1].strip() == "```":
+ lines = lines[:-1]
+ return "\n".join(lines)
+
+
+# ── Header ────────────────────────────────────────────────────────────────────
+st.title("🤖 Multi-Agent Coding Assistant")
+st.caption(
+ "Describe a coding task and watch three specialized AI agents collaborate to produce "
+ "high-quality, reviewed Python code."
+)
+st.divider()
+
+# ── API key handling ──────────────────────────────────────────────────────────
+api_key = os.getenv("MISTRAL_API_KEY", "")
+if not api_key:
+ st.warning(
+ "No `MISTRAL_API_KEY` found in environment. "
+ "Add it to a `.env` file or paste it below.",
+ icon="⚠️",
+ )
+ api_key = st.text_input(
+ "Mistral API Key",
+ type="password",
+ placeholder="Enter your Mistral API key…",
+ )
+
+# ── Task input ────────────────────────────────────────────────────────────────
+task = st.text_area(
+ "Describe your coding task",
+ placeholder=(
+ "e.g. Build a Python function that reads a CSV file and returns "
+ "a dictionary of summary statistics (mean, median, std dev, min, max) "
+ "for each numeric column."
+ ),
+ height=110,
+)
+
+run_button = st.button("🚀 Run Pipeline", type="primary", disabled=not (api_key and task))
+
+# ── Pipeline ──────────────────────────────────────────────────────────────────
+if run_button:
+ if not api_key:
+ st.error("Please provide your Mistral API key.")
+ st.stop()
+ if not task.strip():
+ st.error("Please describe a coding task.")
+ st.stop()
+
+ llm = create_llm(api_key)
+
+ # Step 1 — Planner
+ st.markdown(
+ '<span class="agent-badge badge-planner">🗺️ Planner Agent</span>',
+ unsafe_allow_html=True,
+ )
+ with st.spinner("Planner is breaking down the task…"):
+ plan = run_planner(llm, task)
+ st.session_state["plan"] = plan
+
+ # Step 2 — Coder (first draft)
+ st.markdown(
+ '<span class="agent-badge badge-coder">💻 Coder Agent — First Draft</span>',
+ unsafe_allow_html=True,
+ )
+ with st.spinner("Coder is writing the first draft…"):
+ draft_code = run_coder(llm, plan)
+ st.session_state["draft_code"] = draft_code
+
+ # Step 3 — Reviewer
+ st.markdown(
+ '<span class="agent-badge badge-reviewer">🔍 Reviewer Agent</span>',
+ unsafe_allow_html=True,
+ )
+ with st.spinner("Reviewer is auditing the code…"):
+ review = run_reviewer(llm, draft_code)
+ st.session_state["review"] = review
+
+ # Step 4 — Coder (final version)
+ st.markdown(
+ '<span class="agent-badge badge-final">✨ Coder Agent — Final Version</span>',
+ unsafe_allow_html=True,
+ )
+ with st.spinner("Coder is applying feedback and producing the final version…"):
+ final_code = run_final_coder(llm, plan, draft_code, review)
+ st.session_state["final_code"] = final_code
+
+ st.success("Pipeline complete!", icon="✅")
+ st.divider()
+
+# ── Results ───────────────────────────────────────────────────────────────────
+if all(k in st.session_state for k in ("plan", "draft_code", "review", "final_code")):
+
+ with st.expander("🗺️ Plan", expanded=True):
+ st.markdown(
+ '<span class="agent-badge badge-planner">Planner Agent</span>',
+ unsafe_allow_html=True,
+ )
+ st.markdown(st.session_state["plan"])
+
+ with st.expander("💻 First Draft Code", expanded=False):
+ st.markdown(
+ '<span class="agent-badge badge-coder">Coder Agent</span>',
+ unsafe_allow_html=True,
+ )
+ clean_draft = _strip_code_fences(st.session_state["draft_code"])
+ st.code(clean_draft, language="python")
+
+ with st.expander("🔍 Review Feedback", expanded=False):
+ st.markdown(
+ '<span class="agent-badge badge-reviewer">Reviewer Agent</span>',
+ unsafe_allow_html=True,
+ )
+ st.markdown(st.session_state["review"])
+
+ with st.expander("✨ Final Code", expanded=True):
+ st.markdown(
+ '<span class="agent-badge badge-final">Coder Agent — Improved</span>',
+ unsafe_allow_html=True,
+ )
+ clean_final = _strip_code_fences(st.session_state["final_code"])
+ st.code(clean_final, language="python")
+
+ # Streamlit's st.code already renders a copy icon in the top-right corner of
+ # the code block. The button below provides an explicit fallback with feedback.
+ col1, col2 = st.columns([1, 5])
+ with col1:
+ if st.button("📋 Copy Final Code"):
+ st.session_state["copy_triggered"] = True
+
+ if st.session_state.get("copy_triggered"):
+ st.toast("Use the copy icon on the code block above — browser security "
+ "prevents clipboard access from a button outside the code widget.",
+ icon="📋")
+ # Reset flag so toast only fires once per click
+ st.session_state["copy_triggered"] = False
diff --git a/ai_agents/multi_agent_coding_assistant/assets/demo.png b/ai_agents/multi_agent_coding_assistant/assets/demo.png
new file mode 100644
index 0000000..a2c8858
Binary files /dev/null and b/ai_agents/multi_agent_coding_assistant/assets/demo.png differ
diff --git a/ai_agents/multi_agent_coding_assistant/requirements.txt b/ai_agents/multi_agent_coding_assistant/requirements.txt
new file mode 100644
index 0000000..6e0a954
--- /dev/null
+++ b/ai_agents/multi_agent_coding_assistant/requirements.txt
@@ -0,0 +1,5 @@
+streamlit>=1.32.0
+langchain>=0.2.0
+langchain-mistralai>=0.1.0
+langchain-core>=0.2.0
+python-dotenv>=1.0.0