Advisors

Advisors is an AI‑assisted, multi‑agent app that coordinates domain specialists (entrepreneurship, NFL fantasy, personal finance, legal, software, design, marketing, research, careers, fashion) to produce a clear, actionable plan.

Features

• Multi‑agent “team meeting” to analyze a problem and synthesize a consensus plan
• Advisor Category selector with dynamic presets (leader + specialists, goals/roles)
• Category‑specific agenda placeholders, default questions, and guardrails
• Transparent artifacts: full transcript (.md) and raw messages (.json)
• Streamlit UI for agenda, questions, rules, and team configuration
• Clarity Assistant: auto‑suggests clarifying questions; your answers guide the agents
• Optional PubMed search (via the underlying framework)

Built‑in Categories (editable)

• Medical diagnosis
• Entrepreneur ideas
• NFL fantasy player selection
• Personal finance and investing
• Legal strategy and contracts
• Software architecture and DevOps
• Product design and UX
• Marketing and growth
• Academic research and writing
• Career coaching and hiring
• Fashion

Safety & Scope

• Educational/prototype use only; not medical, legal, or financial advice
• Avoid PHI and confidential data; comply with policies and laws
• Human oversight required; verify claims and consult professionals where applicable

Quickstart

1. Setup

# Install uv (fast Python package manager)
curl -LsSf https://astral.sh/uv/install.sh | sh

# Create and activate venv
uv venv
source .venv/bin/activate

# Install deps
uv pip install -r requirements.txt

2. API key

• UI (recommended): create ./.streamlit/secrets.toml

OPENAI_API_KEY = "sk-..."

• CLI script: create .env (used by medical_consensus.py)

cp .env.example .env
# edit .env and set: OPENAI_API_KEY=sk-...

3. Run UI

uv run streamlit run app.py
# open http://localhost:8501

Using Advisors

1. Choose an Advisor Category (top of the main panel)
2. Describe your case/problem in “Case Description (Agenda)”
3. Optionally click “Suggest Questions” and answer clarifiers
4. Review/adjust “Agenda Questions” and “Rules/Guardrails” (auto‑filled per category)
5. Review the prefilled team (leader + specialists) and edit titles/expertise if desired
6. Click “Run Advisors”, then review tabs: Consensus Summary, Transcript, Raw JSON
7. Saved artifacts: advisor_meetings/<session>.md and <session>.json

Configuration Tips

• Models: Clarifying questions use your selection (e.g., gpt‑5‑nano). Team meeting uses GPT‑4.1‑nano when a gpt‑5* model is selected (Assistants requirement).
• Fast mode: Toggle on to use 1 round and smaller specialist models for lower latency/cost.
• Web search: DuckDuckGo summaries are on by default. Disable for fully offline runs.
• Caching: Use the “Cache outputs” toggle to reuse recent results; turn off when iterating prompts.
• Actionability: Recommendations are enforced to be short, numbered action plans with owners, deadlines, steps, tools, metrics, risks.
• Safety: Web highlights are rendered safely (no HTML). Secrets should live in ./.streamlit/secrets.toml only.

CLI Example

python medical_consensus.py

Edit AGENDA, AGENDA_QUESTIONS, and AGENDA_RULES in the script as needed.

Deploy

• Local: Streamlit run as above
• Container: build a minimal Python image and expose port 8501
• Internal use: protect behind SSO/reverse proxy; store .env securely (not in git)

License

• App code: MIT (adjust as needed)
• Virtual Lab: see license in the upstream repository