Spaces:
Runtime error
Runtime error
File size: 3,053 Bytes
bbb543e | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 | # IBM watsonx Usage Report
## Overview
ClaimCheck AI leverages IBM watsonx as the backbone for claim verification.
The system uses IBM watsonx.ai’s large language models and embedding capabilities to handle:
1. **Text Retrieval (RAG)**
2. **Claim Verification**
3. **Evidence Reranking**
4. *(Optionally)* IBM Speech-to-Text for transcription.
---
## Agents & IBM Integration
### 1. **Transcriber Agent**
- **Goal**: Convert audio from Zoom/phone calls to text.
- **IBM Usage**: Can integrate IBM Speech-to-Text (STT) to provide enterprise-grade transcription accuracy.
- **Inputs**: WAV/MP3/OGG audio file.
- **Outputs**: JSON array of `{ speaker, start, end, text }`.
---
### 2. **Claim Extractor Agent**
- **Goal**: Identify factual statements in the transcript.
- **IBM Usage**: Powered by IBM watsonx.ai LLM (Prompt Lab) to output JSON with `{ claim_text, speaker, start, end }`.
---
### 3. **Retriever Agent**
- **Goal**: Retrieve relevant knowledge base snippets to support or refute claims.
- **IBM Usage**:
- **IBM Text Embeddings**: Generates semantic vector representations of KB snippets and claims.
- **FAISS Vector Search**: Finds top-K relevant evidence from local index.
- **IBM Rerank API**: Orders results by relevance.
- **Benefit**: Reduces false positives in evidence retrieval.
---
### 4. **Verifier Agent**
- **Goal**: Cross-check each claim against retrieved evidence.
- **IBM Usage**: LLM prompt to classify as "Supported", "Contradicted", or "Unverifiable".
- **Output**: Verdicts with reasoning.
---
### 5. **Summarizer Agent**
- **Goal**: Produce final human-readable report.
- **IBM Usage**: LLM summarization model to combine claims, evidence, and verdicts into structured output.
---
## Deployment & Scalability
While deployment is not in the scope of this hackathon, in a production setting:
- **IBM watsonx.ai models** can run in a secure IBM Cloud environment.
- **IBM Orchestrate** could be used to automate claim verification workflows (triggered after each meeting).
- **IBM Code Assistant** could accelerate code integration for enterprise clients.
- **Scalable Vector Store**: IBM Cloud Object Storage could hold large KB datasets for retrieval.
---
## Required Capabilities & Datasets
- **KB Sources**: SLA documents, compliance policies, contractual clauses, product documentation.
- **Integrations**: Zoom/Teams API for call recordings.
- **Agents**: 5 specialized agents as described.
- **IBM Services**:
- Text Embeddings API
- Text Rerank API
- LLM Prompting via Prompt Lab
- (Optional) Speech-to-Text
---
## Why IBM watsonx is Critical
- **Enterprise readiness** — Secure, compliant AI processing.
- **High accuracy retrieval** — Embeddings + Rerank reduce noise.
- **Custom KB adaptability** — Works across industries.
- **Unified ecosystem** — AI models, APIs, and orchestration tools in one platform.
By combining these capabilities, ClaimCheck AI transforms unstructured meeting dialogue into **evidence-backed, compliance-ready reports** in near real-time.
|