Skip to content

Portfolio Projects#

Six end-to-end agent MVPs. Each is a genuinely useful business agent you can fork, extend, and ship this weekend. All run in SWARM_MOCK=true mode with no API key. All have passing tests.

Total: ~78 tests, 6 agents, 0 toy demos.

  • Customer Support Agent

    Multi-turn support agent with order lookup, customer lookup, shipping status, FAQ retrieval, and HITL escalation rules (repeat errors, frustration keywords, irreversible actions over threshold).

    ~$0.01 per conversation · 18 tests · Ch03b + Ch04 + Ch07

  • Code Review Bot

    PR triage: parses the diff, runs 8 security regex checks, detects missing tests, flags high complexity, produces a structured APPROVE / REQUEST_CHANGES / NEEDS_DISCUSSION review.

    ~$0.05 per PR · 18 tests · Ch03b + Ch05 + Ch07

  • Data Analyst Agent

    Natural-language-to-SQL-to-chart pipeline. Ask "top customers by spend last quarter"; get a matplotlib chart, a 2-sentence summary, and the SQL that produced it.

    ~$0.02 per query · 10 tests · Ch03a + Ch03b + Ch05

  • Research Assistant

    Document research with page-level citations. Searches a 10-doc corpus on software engineering history, takes structured notes, synthesizes with inline [1] citations, and runs a grounding checker before returning.

    ~$0.10 per research query · 18 tests · Ch03b + Ch04 + Ch05

  • Knowledge Base RAG

    Hybrid keyword + semantic search over 15+ markdown runbook docs. Answer grounding checker verifies every claim maps to a retrieved chunk. Ships with a realistic SaaS-company knowledge base.

    ~$0.005 per query · 10 tests · Ch04 (+ Vector Memory appendix)

  • Multi-Agent Debate

    Decision support via N debater agents + moderator. Six-persona role library (database expert, performance skeptic, cost analyst, dev experience, risk manager, pragmatist). Consensus detection via Jaccard overlap. Evidence table flags which claims have numeric backing.

    ~$0.15 per decision · 15 tests · Ch04 + Ch06

Running a project#

Clone the repo, install in dev mode, then run any project with a question:

git clone https://github.com/TheAiSingularity/crafting-agentic-swarms.git
cd crafting-agentic-swarms
pip install -e ".[dev]"

# Pick one and run it:
SWARM_MOCK=true python -m projects.customer_support_agent.agent "I want to return order #12345"
SWARM_MOCK=true python -m projects.code_review_bot.agent projects/code_review_bot/sample_diffs/security_issue.diff
SWARM_MOCK=true python -m projects.data_analyst_agent.agent "What are the top 5 customers by total spend?"
SWARM_MOCK=true python -m projects.research_assistant.agent "When did Unix get the pipe operator?"
SWARM_MOCK=true python -m projects.knowledge_base_rag.agent "How do I deploy to staging?"
SWARM_MOCK=true python -m projects.multi_agent_debate.agent "Monolith vs microservices?"

Each project's README.md has the architecture diagram, cost estimate, failure modes, and extension guide for going from MVP to production.

Pick one for your portfolio#

If you finish the course and pick one project to ship, you'll stand out in interviews. Concrete suggestion: pick the project closest to your domain, extend it with two real integrations (real API + real DB), add authentication, put it behind a FastAPI endpoint, deploy to Fly.io or Railway, write a short post-mortem.

That's a portfolio piece. Hiring managers will have opinions about LangChain. They'll respect someone who built the underlying loop.