Kyruus Health wanted to improve patient-to-provider search, which was hindered by checkbox-based interfaces and the need for patients to translate symptoms into medical jargon. These traditional flows often returned irrelevant or incomplete provider results. The friction created decision fatigue and risked higher abandonment. Kyruus Health also needed an experience that matched expectations set by modern conversational GenAI tools. Kyruus Health partnered with Tribe to accelerate a generative AI transformation from proof of concept to general availability. Together they built “Guide,” a conversational search experience that let patients use natural language queries and received provider matches in seconds. The system routed queries through an API Gateway to a conversation service using Claude 3.5 Sonnet on Amazon Bedrock, extracted structured clinical parameters, retrieved candidates via Amazon OpenSearch, and generated clear conversational recommendations. Tribe also supported testing infrastructure, hallucination mitigation, and model validation while embedding with Kyruus Health’s engineering team. The partnership shortened the delivery timeline versus internal expectations. Instead of taking 4–6 months, the team reached production in two months. With Guide in production, Kyruus Health reported improved conversion for members scheduling appointments and faster, more reliable discovery of appropriate care options. The new experience also supported higher provider match satisfaction and faster appointment scheduling.

Follett Software wanted to modernize its Destiny® Library Manager from a static catalog into an intelligent assistant for K-12 librarians. Librarians managed collections that sometimes exceeded 100,000+ titles using tools that could be complex, costly, and less intuitive. Although Follett had an existing AI team and strong ML foundations, it struggled to move beyond early LLM experimentation into a usable product experience. The company needed a clearer path to simplify the interface while delivering actionable insights that improved librarian workflows and ROI. A prototype AI-powered chat interface was built to let librarians interact with their collections using natural language. The solution used a custom text-to-SQL approach with multiple LLM layers (including Anthropic Claude models) to translate questions into reliable SQL queries. It followed a retrieval-augmented generation (RAG) architecture to return table results and support filtering and refinement. The system was designed to handle typos and understand library-specific concepts like the Dewey Decimal System, and it was implemented as a containerized cloud application on AWS Bedrock with an automated deployment pipeline. By the end of the engagement, Follett had a working prototype that enabled natural-language querying of school library collections. The interface returned clear answers and surfaced purchase and weeding recommendations, improving decision-making and creating a potential upsell pathway. Follett’s engineering team was also upskilled with patterns and frameworks to support and extend the solution internally. The work positioned Follett to scale AI as a platform-level capability across its broader portfolio of sellable products and suites.

Keyloop’s aftersales operations relied on a manual, inefficient process for scheduling technicians. Workshop administrators built daily schedules from scratch despite constantly changing variables such as job types, technician skills, and bay availability. The lack of automation created administrative burden and constrained growth. It also negatively affected customer experience and left revenue unrealized. Keyloop partnered with Tribe AI to build an intelligent scheduling engine for its Service Hub product. The system incorporated real-world constraints like technician skills and job priority to automatically generate optimized daily schedules. Advisors started each day with a data-informed plan they could fine-tune rather than a blank slate. The approach included a technician schedule optimizer and a human-in-the-loop review process to support trust, accuracy, and compliance. The AI-driven scheduling capability reduced the need for workshop administrators to create schedules from scratch each morning. It improved workshop throughput and job prioritization by producing a ready-to-use daily schedule. The change supported higher technician utilization and faster service turnaround for dealerships. Keyloop also established a foundation to expand AI into additional initiatives, including work to accelerate OEM certifications through automation and structured schema mapping.

Avalon managed over 60 complex, frequently updated policy documents used to determine patient eligibility for diagnostic tests. The prior authorization process was manual and time-consuming, requiring staff and providers to sift through extensive documentation. This created operational inefficiencies and increased the risk of delays or errors in patient care. Avalon partnered with Tribe to build a customized proof of concept using large language models to extract key information from policy documents and generate medically accurate prior authorization questions. The team used Claude 3 Opus to parse policy language, identify relevant coverage sections, and produce structured clinical questions for human reviewers. The PoC also supported PDF uploads, model selection, auto-generated test lists with manual adjustment, and a feedback loop to refine outputs with human-in-the-loop oversight. The PoC pilot achieved 100% precision in producing clinically accurate follow-up questions for human reviewers and delivered 83% recall across the policy documents tested, exceeding internal benchmarks. These results supported fewer eligibility assessment errors and faster review times. Avalon also shifted its policy documentation review cadence from annual to monthly cycles and planned to expand beyond the initial four test policies.

Ataccama accelerated its use of AI but struggled to maintain and leverage quality data at scale. Department-level experiments and ad hoc automation created isolated efficiency gains while fragmenting the company’s trusted data foundation. The impact showed up in operations and go-to-market work, including 180 person-days per year lost to manual RFP responses and 80–150 hours per deal spent on repetitive sales preparation. Support operations also suffered, with 60–70% of tickets stuck in manual triage and information scattered across Slack, Notion, Jira, and Google Drive. Ataccama partnered with Tribe to operationalize data trust inside its own business through an AI Acceleration Framework. The framework applied governance, observability, and data quality principles to identify and automate high-impact repetitive work while preserving oversight. It implemented automated RFP responses and content generation using governed internal data, and it introduced Level 1 support triage powered by classification and enrichment models trained on trusted sources. An AI Program Office was also established to centralize governance, track ROI, and ensure consistent and compliant AI deployment across teams. In parallel, Ataccama extended its data trust foundation externally by creating an enterprise-grade Model Context Protocol (MCP) server. This transformed Ataccama ONE Agent’s 14+ capabilities into AI-ready, governed services with authentication, observability, and scalability for use across tools like Power BI, Claude Desktop, Snowflake Cortex, and developer IDEs. Internally, the company captured measurable gains by reducing manual effort and improving responsiveness in sales and support workflows. The dual approach positioned Ataccama as a cross-platform AI Trust Layer while enabling AI systems to query trusted enterprise data more directly.

MyFitnessPal wanted to improve its user experience by incorporating generative AI into its product suite. The team needed additional capacity to scope, prototype, and iterate on innovative features. A key goal was increasing engagement by enabling users to log food conversationally instead of searching and typing each item. The desired experience required accurately capturing items, quantities, and even brands or packaged foods in a single step. MyFitnessPal brought in an external team to extend its product organization and help kickstart its AI product roadmap. The team built a functional demo in a sprint for a prototype called Voice Log that supported logging multiple items in one interaction. The flow let members enable the feature in-app, grant microphone permissions, speak meals in everyday language, and receive best-match suggestions from the database. The implementation used a full-stack, cloud-based approach integrated with MyFitnessPal’s existing environment. The project delivered a working Voice Log demo that streamlined multi-item food logging into a single conversational interaction. It also helped MyFitnessPal accelerate its AI product roadmap and informed how the company approached experimentation. Leadership was influenced to invest further in a culture of experimentation and to create a dedicated prototyping environment. MyFitnessPal positioned itself to identify and develop additional AI-powered features to strengthen its leadership in health and fitness.

A global consulting firm specializing in corporate turnarounds and bankruptcy restructuring needed a faster way to analyze complex vendor ecosystems during diligence. The existing approach required weeks of manual effort to classify thousands of vendors into a multi-level taxonomy. Limited and inconsistent vendor information made the work error-prone and difficult to scale. The slow, manual process delayed downstream diligence deliverables and insight delivery. Tribe AI partnered with the firm to design and implement an AI-powered vendor classification engine inside the firm’s proprietary due diligence platform. The system used OpenAI GPT-4o and GPT-3.5-turbo with structured outputs, combined with historical classification patterns, deterministic business rules, and confidence scoring. It ingested raw vendor data, generated category predictions, applied rules-based mappings, and routed low-confidence cases for human review. A feedback loop captured human corrections to improve classifications over time. The firm categorized 18,000+ vendors in a live engagement with 97% accuracy, validated by subject matter experts. The new workflow reduced process time from 9–12 days to 45 minutes. The engine was used across the firm’s Private Equity and Turnaround & Restructuring service lines. It was actively deployed on hundreds of thousands of vendors and set as the standard going forward.

A global consulting firm’s investment diligence process relied on “outside-in” research across public sources to assess risks, opportunities, and trends for target companies. The work was constrained by manual workflows, with analysts spending days or weeks gathering and stitching together fragmented signals from news, filings, job postings, and reviews. This created delays, risked missed signals, and led to inconsistencies across projects. The firm aimed to cut execution time down to 3 days by using an AI-driven approach. The firm partnered with Tribe AI to build an AI-powered research assistant embedded in its proprietary due diligence platform. The assistant aggregated data from public and proprietary sources and extracted relevant signals such as financials, employee rosters, salary data, leadership changes, and hiring trends. It presented insights in a searchable, curated dashboard with source citations and outputs compatible with Excel and PowerPoint. The system used NLP-based extraction, relevance ranking, and feedback loops to improve results over time. The implementation accelerated outside-in due diligence and improved consistency across projects. Outside-in research time was reduced from 7–10 days to under 1 day. Signal coverage increased by 30%, helping capture insights that manual workflows would have missed. Manual analyst hours were reduced by 40% per diligence project, and reporting was accelerated by 2–3 days to enable earlier decision-making.

A hardware and software development company set out to build the industry’s first commercial utility Ground Penetrating Radar (GPR) system for safely verifying underground utilities before digging. Standard site-assessment procedures still produced frequent inaccuracies, causing unplanned disruptions, delays, and financial liability for contractors. The team needed higher accuracy and portability than existing offerings and believed AI could materially improve detection performance. They also wanted the work to support patentable intellectual property. An end-to-end prototype was built to analyze scans from a custom GPR device and detect and locate underground objects. The system used AWS Elastic Kubernetes Service (EKS) for workload management, Python for machine learning, and React/Node for the application interface and backend. IMU data was integrated to correct scans and add context about ground conditions, while automated ingestion and preprocessing (e.g., de-wow, normalization, denoising, contrast enhancement, and data translation) prepared data for model inference. A tablet-based UI allowed users to run inference locally and review results as points or bounding boxes with confidence scores, with depth estimation informed by user-specified substrate type. By the end of the initial engagement, the team received a highly accurate (~90%+) prototype that could detect and locate underground infrastructure from the custom GPR scans. The handheld device concept targeted quick field adoption with an expected learning curve of about five minutes and produced a scan for a 10’ x 10’ area in about three minutes. The radar capability extended up to 5 feet deep, and a visual map was generated within about 30 seconds after scan completion. The prototype’s preprocessing and system design contributed to a successful patent application and helped attract investor interest and new funding to move the product into production.