Client profile
Resolve Dynamics is a fractional Chief AI Officer firm helping mid-sized businesses ($10M to $50M in revenue) implement practical AI solutions.
The challenge
A Resolve Dynamics client needed AI assistants powered by a custom fine-tuned model so that it would truly understand their business, not generic chatbots, but smart agents trained on their specific documents, policies, people, and history.
- •One client had roughly 8,000 documents to power their chat interface, far beyond what standard RAG context windows could effectively process.
- •Another client had indexed 1,500 files into 122,000 chunks, with a history of 18,000+ completed projects representing untapped institutional knowledge.
- •Existing proposal generation was template-based and rigid, unable to draw on organizational context.
- •Field reports required manual, time-consuming effort with no automation.
- •No compliance-grade tracking existed for AI-generated decisions, a critical gap for clients in regulated industries like healthcare and finance.
Resolve Dynamics needed a partner who could fine-tune a model to power the custom AI solution without needing to hire an ML team, subject matter experts to contrive a "fake workflow," or a team to train and clean their data, without requiring a massive budget.
The solution
Empromptu AI’s platform builds an AI application that a team and its users actually use, and captures the production output to fine-tune custom models. Resolve Dynamics didn’t have to buy data from a data labeler or hire a team to fine-tune a custom model.
Empromptu AI built a custom AI chatbot purpose-built for Resolve Dynamics’ customer’s operations, and used the output to fine-tune a Qwen model. The solution combines a fine-tuned language model with a comprehensive knowledge base, enabling Resolve Dynamics’ customer to deploy business-specific AI assistants that go far beyond generic chat tools, with up to 98% accuracy.
Key capabilities include:
- •Fine-tuned Qwen model: A custom-trained AI optimized for the customer’s specific use cases and language.
- •Comprehensive knowledge base: All company documents uploaded to the backend: employee resumes, company handbooks, project histories, onboarding materials, policies, and proposals.
- •Source citations: Every AI response includes citations linking back to the exact source document, enabling verification and trust.
- •Configurable strictness levels: From strict mode (only referencing uploaded materials) to open mode (generating new ideas and insights), tunable per use case.
- •AI decision log: A compliance-grade audit trail tracking every AI decision, its rationale, and supporting evidence, critical for regulated industries.
- •Flexible deployment: Deployable on Resolve Dynamics’ own infrastructure or through deployment partners, with a path to on-premises installation.
How it works
The workflow is intentionally simple on the front end, with sophisticated AI doing the heavy lifting behind the scenes:
- •A user submits a query via a clean, simple chat interface.
- •The AI searches the knowledge base using both graph similarity and vector similarity, going well beyond standard keyword or single-vector retrieval to surface the most contextually relevant information.
- •Pre-extraction and pre-summarization techniques process large document sets that would overflow standard context windows.
- •The AI generates a response with citations, showing exactly which documents informed the answer.
- •Optional human review is supported, with the AI decision log capturing all outputs for compliance and oversight.
Real examples from the live demo:
- •"Tell me about Mark’s resume and what project types he might be good for": The system returned Mark’s 30 years of experience in project management and construction, his capabilities, and matching project types, with sources.
- •"What is our leave policy?": The system retrieved the relevant section from the company handbook and surfaced the guidance with a recommendation to verify against the source.
Integrations
- •SharePoint: Document library ingestion (1,800+ files, 122,000+ chunks indexed).
- •Slack: Primary collaboration and communication channel.
- •Microsoft Teams to Slack sync: For clients preferring Teams-based workflows.
- •Nvidia infrastructure: GPU capacity for model training and scaling.
- •On-premises servers: Planned deployment path for Resolve Dynamics’ own infrastructure.
Results
The solution launched in its initial deployment phase, with Resolve Dynamics’ team and end users beginning testing and feedback cycles immediately. While formal ROI metrics are still being captured, the early indicators are strong:
- •A fine-tuned model in less than 30 days.
- •1,800+ historical projects now accessible as a queryable knowledge asset for proposal generation and field reports.
- •Proposal generation upgraded from rigid templates to context-aware, data-informed drafts.
- •Compliance-ready AI decision logging, enabling Resolve Dynamics to serve clients in healthcare and finance with confidence.
- •A scalable model: the fine-tuned Qwen architecture is designed to expand horizontally across Resolve Dynamics’ entire client portfolio.
“We’re rocking and rolling. We’ve deployed 12 projects end to end in recent weeks... we’re outpacing our revenue goals. There’s so many opportunities.”
Marty Reed
Resolve Dynamics
