|
| 1 | +--- |
| 2 | +title: BargAIner |
| 3 | +description: Real-time procurement negotiation assistant that listens to vendor calls, flags risky claims, and suggests data-backed counter-arguments. |
| 4 | +authorIds: |
| 5 | + - max-schallermayer |
| 6 | + - nikolai-khalatiani |
| 7 | + - anton-schedel |
| 8 | + - oskar-kraak |
| 9 | +categories: |
| 10 | + - speech-to-text |
| 11 | + - agents |
| 12 | +isFeatured: false |
| 13 | +date: "2025-01-20" |
| 14 | +image: /images/bargainer.png |
| 15 | +repoUrl: https://github.com/bargAInerTUMAI/bargAIner |
| 16 | +videoUrl: https://drive.google.com/file/d/1PBAoM1JAaKA4B0a8TtIj3PJG-ObuC9Sl/view?usp=sharing |
| 17 | +--- |
| 18 | + |
| 19 | +# BargAIner |
| 20 | + |
| 21 | +## Overview |
| 22 | + |
| 23 | +BargAIner is a desktop application providing real-time AI assistance during procurement negotiations—built for Software Migration & Cloud Modernization deals. It listens to vendor conversations, detects claims that could negatively impact the buyer (pricing, timelines, scope), and generates instant counter-arguments backed by data. |
| 24 | + |
| 25 | +ElevenLabs is used for **real-time Speech-to-Text via Scribe**, streaming transcripts into the UI and agent pipeline. |
| 26 | + |
| 27 | +## Key Features |
| 28 | + |
| 29 | +- **Real-time Speech-to-Text (ElevenLabs Scribe):** Live transcription with VAD-based segmentation |
| 30 | +- **Mic + system audio capture:** Captures both sides of the conversation |
| 31 | +- **Claim detection:** Flags risky vendor claims (rates, scope exclusions, timeline pressure) |
| 32 | +- **Fact verification:** Checks an internal knowledge base + web search benchmarks |
| 33 | +- **Counter-argument generation:** Produces data-backed responses in real time |
| 34 | +- **Wrap-up + action items:** Detects summaries and extracts next steps; exports to PDF |
| 35 | +- **Post-negotiation feedback:** Scoring, tactics assessment, and improvement suggestions |
| 36 | + |
| 37 | +## How It Works |
| 38 | + |
| 39 | +1. **Electron frontend** captures microphone + system audio. |
| 40 | +2. Audio is streamed to **ElevenLabs Scribe via WebSocket** for live transcription. |
| 41 | +3. Transcripts are sent to a backend agent loop (Cerebras) which: |
| 42 | + - detects triggers (scope, pricing, staffing rates, timeline), |
| 43 | + - verifies claims against local documents (knowledge base), |
| 44 | + - pulls benchmarks via web search (Tavily), |
| 45 | + - generates counter-arguments and action items. |
| 46 | +4. The UI shows real-time transcripts, suggestions, and wrap-up/action items. |
| 47 | + |
| 48 | +## Technologies Used |
| 49 | + |
| 50 | +- **ElevenLabs Scribe (Speech-to-Text, WebSocket streaming)** |
| 51 | +- Electron, React, TypeScript, Node.js, Express |
| 52 | +- Cerebras (fast inference) |
| 53 | +- Tavily (web search) |
| 54 | +- Vercel AI SDK (agent framework) |
| 55 | + |
| 56 | +## Getting Started |
| 57 | + |
| 58 | +Full setup instructions are in the repository README: |
| 59 | +https://github.com/bargAInerTUMAI/bargAIner |
| 60 | + |
| 61 | +## Demo |
| 62 | + |
| 63 | +See the repository for instructions, or checkout our live demo at: |
| 64 | +https://drive.google.com/file/d/1PBAoM1JAaKA4B0a8TtIj3PJG-ObuC9Sl/view?usp=sharing |
| 65 | + |
| 66 | +## Future Plans |
| 67 | + |
| 68 | +- Add more negotiation “personas” and configurable playbooks |
| 69 | +- Improve benchmark sourcing and citations for claims |
| 70 | +- Ship a hosted demo / packaged releases for easier testing |
| 71 | + |
| 72 | +## Acknowledgments |
| 73 | + |
| 74 | +Built at the TUM.ai Hackathon by: |
| 75 | +- Max Schallermayer |
| 76 | +- Nikolai Khalatiani |
| 77 | +- Anton Schedel |
| 78 | +- Oskar Kraak |
0 commit comments