Prakhar Saxena

Lead a ~30-person AI organization—principal and senior software engineers, architects, senior business analysts, and a product owner—as a leader of leaders, with team leaders running the platform teams day to day. Own strategy across three enterprise AI programs built from zero: a multi-tenant document-interrogation platform whose largest client is UWM's flagship underwriting product, BOLT; an agentic, spec-driven software development platform; and a company-wide AI Enablement program for all ~9,000 team members.

• Proposed a company-wide AI Enablement program directly to the SVP and all of the VPs reporting to him—taking AI enablement beyond IT to all ~9,000 team members—and won the mandate to build it from zero; now partner with three VPs to deliver it.
• Authored the program's operating model around four questions—What to use? When and why to use it? How to use it? Is it working?—and organized the team to answer them through three platforms run in parallel: Evaluation, Agent, and Personal AI.
• Building an in-house Evaluation Platform that scores models, skills, agents, and agent harnesses on accuracy, cost, latency, adoption, and productivity—measured from system data such as Jira cycle time and tool usage, not self-reported surveys.
• Defining UWM's standards for how agents are built—design, security, guardrails, integrations, and observability—through the Agent Platform.
• Struck a deal with the business over who builds agents: business teams may build their own, but each must pass a hard review by an independent team—about five days on average—before full use, meeting business demand without giving up control.
• Lead the Personal AI Platform—the AI tools team members use every day: Claude Code, GitHub Copilot, and Cursor for ~1,400 developers, and Microsoft 365 Copilot for the other ~7,600 team members, now at ~70% adoption and climbing.
• Rolling out training through UWM's existing IT training team rather than building a separate one—training the trainers and supporting them closely; weekly sessions now draw ~40 team members.
• Created company-wide skill and prompt catalogs, plus a one-stop shop where team members describe their use case and are pointed to the right skill or prompt, organized by role.
• Founding, voting member of UWM's AI Standards Council, which works with Legal to interpret the laws governing AI and sets the AI standards the whole company follows; partner closely with Security, Legal, and Compliance on AI policy.

• Founded a second platform from zero in 2026—an agentic, spec-driven development platform that ingests specification documents and autonomously produces production-ready code—extending the build-from-zero playbook from business automation into software engineering.
• Scaled the platform from pilot to near-universal adoption across UWM's IT organization, generating 1,000+ pull requests within a single month.
• Cut Jira cycle time ~40% for pilot teams and ~30% across IT at full rollout—about 870K hours, or roughly 420 developers' worth of delivery capacity, a year.
• Drove the shift to AI-written code: nearly all new code is now generated with AI, with developers reviewing and refining rather than hand-writing it.
• Built resilient services and quality gates requiring 100% test coverage and an 80% mutation score, so quality rests on proven engineering practice rather than trust in the model.
• Architected the platform model-agnostic—interchangeable Google Gemini, Anthropic Claude, and self-hosted open-weight models served via vLLM—eliminating vendor lock-in and routing workloads on cost and performance.
• Now breaking the platform into modular pieces that can each be tuned and controlled independently.

• Conceived and built UWM's document-interrogation automation platform from the ground up—now the largest enterprise deployment of Google LLMs at the company—converting manual, document-heavy underwriting reviews into real-time, AI-driven decision support.
• Identified the opportunity as an individual contributor and authored the founding business proposal to VP/AVP stakeholders, securing executive sponsorship to stand up the team from zero—then hired, structured, and matured the org built from there.
• Delivered ~$4.1M in annualized savings ($346K/month) through strategically targeted platform optimizations, with additional millions realized through loan-quality and underwriter-efficiency gains.
• Returned over a million underwriter hours per year to the business—10 hours/week across ~2,000 underwriters—the equivalent of roughly 500 full-time work-years of reclaimed capacity.
• Expanded the platform from a single underwriting use case into a multi-tenant service supporting client products enterprise-wide, handling any document-based review requiring complex question answering.
• Architected ML solutions using Google Gemini, custom-trained models, and embeddings to automate Title Review, Assets, Purchase Agreements, Pay Stubs, Credit Reports, and Appraisals.
• Working hands-on with an architect on a proof of concept to reduce dependence on Google's hosted Gemini APIs—benchmarking self-hosted, open-weight Gemma 4 against Gemini on the document-interrogation pipelines, prompted in part by Evaluation Platform findings.
• Targeting near-zero marginal inference cost by shifting AI spend from OpEx to CapEx, and an end to the retraining and model-migration work forced whenever a vendor retires a model.

• Building the AI Enablement team from zero—roughly eight people so far, with senior developers and architects leading the work, and still expanding.
• Own people leadership for the ~30-person organization—hiring, career development, and performance management for team leaders, principal/senior engineers, architects, and business analysts—standing up each new team while keeping the existing platforms running.
• Step into the PM, Product Owner, and Scrum Master roles whenever a new program starts, until it can staff them—now for AI Enablement, as earlier for both platforms.

Member of the BOLT Architects team; led multiple product- and organization-critical initiatives.

• Drove currency-management efforts that cut technical debt and eliminated security vulnerabilities across multiple BOLT services.
• Led a cohort of summer team members through a full-stack microservices project (React + .NET 8), achieving 75% retention.
• Acted as technical liaison to vendor partners (Qarik/Egen), owning environment setup and integration; mentored developers across frameworks and tooling.

• Administered an on-premises Confluent Kafka platform at 99.99% availability, processing billions of messages daily across the organization's intranet.
• Supported 1,900+ technology professionals integrating with Kafka, Redis, and Qlik; ran the Redis distributed cache platform at ~4,000 operations/second.
• Pioneered open-source Kafka adoption to cut vendor dependency and migrated Kafka connectors into Kubernetes via custom Python tooling.

• Sole maintainer of a full-stack trading application ecosystem at 99.999% availability within an AWS Kubernetes cluster.
• Built Java/Spring gateways bridging legacy trading frameworks to NASDAQ's consolidated Fusion platform; introduced Scala/Java code generators driven by a domain DDL file system.
• Modernized the ecosystem end-to-end, migrating 1990s-era trading software to current Scala (2.8, 3.0) and Java (11, 16) with zero uptime loss during the transition.
• Delivered back-end services integrating multiple international NASDAQ exchanges through cross-border collaboration.

Proposed, staffed, and led the effort to port Quill—a compile-time query library for Scala—onto Scala 3. Unpaid open-source work carried concurrently with the NASDAQ role, in close collaboration with the library's lead maintainer, who was also my mentor at NASDAQ. Served as my Drexel senior capstone. The resulting library, ProtoQuill, became the official Scala 3 implementation of Quill.

• Originated the project: identified the Scala 2 → Scala 3 port as an effort I could own end to end, and pitched it in parallel to the library's lead maintainer and my faculty advisor to establish it as a sponsored senior capstone.
• Recruited and led a team of 5–6 developers sourced through my Drexel network and their extended connections—setting technical direction, dividing ownership of the AST surface, and running delivery through Jira and GitHub.
• Built the lifters and unlifters that translate Scala 3 syntax tree elements into Quill's internal query AST, covering core query constructs including select, filter, and order.
• Designed the compiler-level pipeline end to end: read Scala syntax tree elements at compile time, convert them into a custom query AST, then render that AST into whichever target query language the caller selected—preserving Quill's compile-time query generation on an entirely new metaprogramming foundation.
• Developed against Scala 3 Milestone and Release Candidate builds before the language was stable, working without a settled API surface and surfacing compiler-level regressions in the process.
• Built the team's quality infrastructure from scratch—no testing library existed for Scala 3 at the time, so established review standards and verification processes in place of tooling that did not yet exist.
• NASDAQ ran Quill in production, making the port directly relevant to the firm; portions of the work were approved on company time, with the balance carried after hours alongside the lead maintainer.
• Contributed to a library that reached production release: ProtoQuill is published to Maven Central as the quill-*_3 artifacts and is now maintained under the ZIO organization as Quill's Scala 3 line.

Drexel co-op, secured through a full external application and interview process and run as a full-time role. One of two people in the company's entire Strategy & Planning function, partnered directly with the VP of Strategy & Planning; analysis was consumed by the CEO, Co-Founder, and CTO.

• Rebuilt the company's competitive intelligence pipeline after a legal notice ended its reliance on web scraping—migrated the data supply to sanctioned App Annie and SimilarWeb API integrations, restoring the intelligence feed on a compliant footing.
• Produced recurring competitor analysis—traffic, visits, and revenue growth—and presented findings and my own read of them directly to the CEO, Co-Founder, CTO, and VP of Strategy & Planning.
• Owned analysis and reporting of the company's annual revenue data.
• Built visualizations and reporting in Tableau and ChartIO on top of Google BigQuery, Google Cloud, and Sheets/Excel; consistently recognized by the executive team for the clarity of the visuals and the quality of the insights.
• Cut query load by 100 GB per execution cycle through query optimization and deliberate visualization design, producing meaningful cost savings.

Drexel co-op on the front-office technology team, building directly for the firm's traders.

• Designed, built, and shipped to production a web application giving the firm round-the-clock visibility into the status of critical production-support jobs.
• Implemented RESTful APIs in .NET with Entity Framework, evaluating and tuning the underlying SQL Server queries.
• Delivered the user interface in Angular, the firm's front-end standard at the time.
• Independently prototyped in React and shipped the firm's first React application to production, introducing the framework to the organization.
• Gathered requirements directly from traders and worked with the team to establish shared engineering practices; kept current through in-house and external coursework.

Led a five-person team on an indoor navigation mobile application, carried alongside a 20-credit course load as an Honors College student pursuing a CS major with dual concentrations in AI and Cyber Security and a Data Science minor.

• Led a team of five through design, build, and release of an indoor positioning and navigation mobile application.
• Engineered router-based location tracking to function inside buildings, where GPS accuracy degrades.
• Integrated the Google Maps API for precise location tracking and the Firebase database environment for seamless logins and data storage.
• Shipped the application for students and faculty locating rooms, offices, and other destinations within campus buildings.

Built during secondary school—the first substantial system I designed, built, and shipped end to end for a real user base.

• Led development of a comprehensive library management system for the school library, from requirements through deployment.
• Built the system in Java, using the NetBeans IDE to develop an intuitive user interface for library staff.
• Implemented a SQL backend for efficient and secure management of the library's catalog and lending records.