Band of Agents Hackathon · 2026
Multi-agent review response that earns the reaction it deserves.
Six specialized agents, coordinated live in Band, turn unanswered customer reviews into responses worth a standing ovation.
Why this matters
of customers expect a reply to a negative review within a week.
Most never get one. The ones that do are often generic, slow, or both.
At scale, a thoughtful response isn't a writing problem — it's a coordination problem.
The cast
Coordinated entirely inside Band — no hidden handoffs, no black-box routing. Every message between agents is one judges can read.
Polls connected sources and surfaces new reviews into the Band room as they arrive.
No LLMClassifies sentiment and urgency, then routes the review to the right next step.
FeatherlessEnriches the review with brand voice and business context before drafting begins.
FeatherlessGenerates the candidate response — and rewrites it when QA sends it back.
DeepSeek-V4Scores every draft against tone and quality. Approves it — or loops it back, up to twice.
DeepSeek-V4Catches legal threats, safety issues, and media mentions — holds them for a human, always.
FeatherlessThe coordination showpiece
A draft that misses the mark doesn't get published — it gets sent back, revised, and reviewed again. In the Band room, judges watch every step.
“Every agent message is a real Band message. No hidden handoffs. No black-box coordination.”
Coverage
Every branch of the pipeline — publish, revise, and escalate — exercised and verified before the live demo.
Full pipeline, single pass, QA approves on the first draft.
Brand-voice research is critical to getting the tone right.
Research enrichment is what decides the correct routing outcome.
QA rejects v1 as too verbose. Drafter revises. QA approves v2.
Drafting is skipped entirely. Held for human review, no auto-publish.
Immediate critical hold — severity HIGH, no exceptions.
Comms-team alert fired. No response goes out automatically.
Research and Drafter together navigate genuinely mixed sentiment.
Under the hood
Shared room, agent handles, visible message routing for all six agents.
Powers Triage, Research, and Escalation — cost-efficient inference.
Powers Drafter and QA — premium reasoning for generation and scoring.
A controlled inject endpoint for precise, repeatable live demo scenarios.
# clone and install git clone git@github.com:resilientbeast/review-response-system.git cd review-response-system uv sync # configure cp .env.example .env cp agent_config.yaml.example agent_config.yaml # add BAND_API_KEY, OPENAI_API_KEY, etc. # run the agent pipeline uv run python run_all.py # trigger seed_04 — the QA revision loop showpiece curl -X POST http://localhost:8002/inject \ -H "Content-Type: application/json" \ -d '{ "platform": "tripadvisor", "business_id": "loc_demo", "review": { "text": "Worst experience of my life. The manager was completely dismissive when I raised my concerns. The food was inedible and no one seemed to care. I will be leaving reviews everywhere I can.", "rating": 1, "author": "Sarah K.", "url": "http://demo.platform", "language": "en" } }' # full setup incl. dashboard + ingestion: see Judges Overview