AI product development for professional services, built for PE portfolio value creation
AI product development for professional services, built for PE portfolio value creation
Rosenblatt is an AI product development firm that builds AI-first MVPs for professional services companies. We specialize in the language intensive knowledge work that defines these industries, automating routine operations through multi agent systems, retrieval augmented generation, tool calling integrations, and computer vision for document processing.
Over the past two years, we have delivered 10 AI products for clients ranging from early stage startups to mid market companies backed by institutional PE. Our most advanced engagement, ProService Hawaii, a Silver Lake portfolio company, demonstrates what a PE backed AI transformation looks like in practice.
Why TCV: TCV's portfolio includes several professional services and services adjacent companies, including Employment Hero, Perceptyx, OneSource Virtual, and Watermark, where AI can drive measurable operational leverage. With our PE portfolio experience through Silver Lake and our focus on the professional services vertical, we believe Rosenblatt is well positioned to serve as an AI engineering partner across TCV's portfolio.
What we're proposing: A collaborative exploration, starting with one to two portfolio companies where AI can have the most immediate impact, to demonstrate our approach, build trust, and expand from there.
Rosenblatt builds production AI systems, not prototypes. Our core capabilities map directly to the workflows professional services companies depend on.
AI agents that take action, not just answer questions. At ProService, our Fridai agent pre fills 80% of new client onboarding data by orchestrating across multiple systems and escalates to humans when confidence is low. Built on Claude Sonnet with advanced tool calling and human in the loop escalation.
Most RAG systems retrieve chunks of text. Ours build structured ontologies over client data, enabling precise multi hop reasoning across rapidly growing, diverse knowledge bases. We use Claude Haiku for lightweight ontology construction at scale. Deployed at ProService (call center and onboarding document retrieval) and Rilevera (cyber threat intelligence).
Extracting structured data from messy real world documents. For Campaign Finance Reports, we reduced data extraction time from 30 minutes to 5 minutes using GPT-4o with custom image preprocessing for handwritten documents. For UScope, we built automated property damage photo captioning.
When off the shelf models are not enough, we fine tune. For Branch, we built a three stage pipeline (sentence embedding, relevancy classifier filter, semantic similarity SLM) achieving 95% accuracy matching political candidate quotes to campaign topic categories. For Salient, we trained CPG sales forecasting models using META prophet models.
AI opportunity map, prioritized roadmap, and architecture plan.
Production ready MVP targeting a specific business metric.
AI product development integrated with client team.
On site or virtual deep dive with product and engineering leaders. We leave with two to three high impact AI use cases.
Comprehensive project plan with JIRA tickets and architecture diagrams, ready to execute by our agent team.
Through human-agent collaboration, we deliver enterprise ready MVPs in under four weeks.
Deploy, measure KPIs, iterate based on real user feedback, map user behaviour to automated testing to ensure continued agent team velocity.
Successful pilots lead to broader engagement. ProService doubled their investment after our first successful pilot.
Since our founding in 2024, our engineers have worked exclusively with AI-first coding editors, shipping higher quality code faster than traditional development workflows allow.
We identified early that foundation models routinely overlook security. We adopted SOC-2 practices so every line of code meets enterprise standards from day one.
Today we deploy teams of AI agents guided by experienced AI team leads, combining security rigor with cutting-edge automation to keep our delivery velocity best in class.
Challenge: New client onboarding required multiple calls and manual data entry, causing delays and churn.
What we built: A multi agent onboarding system using Claude Sonnet that pre fills 80% of client data from sales call transcripts, with human in the loop escalation when confidence is low.
Challenge: Customers complained about incorrect information or guidance by call center employees, lowering first call resolution rates.
What we built: A call center copilot using Claude Sonnet, integrated with Snowflake via GraphRAG and launched in three weeks to support frontline agents in real time.
Challenge: CPG sales data collected by weekly delivery information is sparse and hard to fit with linear models.
What we built: Trained grouped prophet models to allow hierarchical forecasting across stores, brands, and products.
Challenge: Match relevant political candidate quotes to campaign topics from large unstructured web corpora with high precision.
What we built: Custom three stage model pipeline: sentence embedding, relevancy classifier, and SLM topic matcher trained from scratch.
Challenge: Handwritten campaign finance documents required 30 minutes of manual data extraction per form.
What we built: Computer vision pipeline using GPT-4o with custom image preprocessing for handwritten field extraction.
Challenge: Early stage threat intelligence company needed a working product to raise seed funding.
What we built: Cyber threat detection agent powered by Claude Sonnet using GraphRAG with Claude Haiku for ontology building, plus Text2SQL tool calling.
Less than 5 years from undergrad to Director of AI Engineering at a Silver Lake portco, inserted by their Operating Executive of Data and AI. Second-time founder, Techstars '23 and AWS Accelerator '22 alumni. 10x'd revenue from year 1 to 2. Led production AI training and deployments in multi-agent systems, image search and captioning, learning to rank, and media mix modeling at both startups and enterprise alike.
···
Backgrounds from Johns Hopkins, UVA, NYU, and Georgia Tech. Prior experience at PwC, Deloitte, Amazon, and the US Air Force.
Not just a collection of system prompts, but well harnessed autonomous agents, across multiple providers, designed to solve complex problems.
This overview is a starting point, not a final answer. We welcome David's team adding context on which portfolio companies are most receptive and where the biggest pain points are.
Identify one to two portfolio companies for an initial pilot. We will come prepared with hypotheses; David's team brings the inside context.
Scoped project with one to two engineers, targeting a specific metric such as resolution rate, processing time, or churn reduction.
Based on pilot results, roll out to additional portfolio companies with a standardized but customized approach.
If this direction is worth exploring, book a working session and we will walk through the portfolio fit, pressure test priorities, and identify the best first pilot together.
Book The Call