I’m Zubin, welcome! I’ve built over 100
useful products for startups, F500s, and the Public Sector. I am a
triple-threat (Engineering, Product and Design) with high agency. I am
currently exploring how agents can empower professionals who want to
maximally leverage their skills and tacit knowledge across organizations
they serve.
ZUBIN RUSTOM WADIA Applied AI| Forward-Deployed Engineering (914) 498-9929 |zwadia@gmail.com | Chappaqua, NY linkedin.com/in/zubinwadia | murch.ai | barkie.ai PROFESSIONAL SUMMARY Forward-deployed productengineer who builds AIproducts the hard way: deep customer discovery to nd the real prob- lem, rigorous evaluation to de ne correctness - even in low-veriability domains where thereis no ground truth - and judicious extension of the core platform to land the world's largest customers without fragmenting it. Directly responsible for 78% of the rst $3M in ARR at an AI document-intelligence startup; earlierarchitected a$2.1B/year government claims platform. APPLIEDAI & ENGINEERING Applied AI:long-running autonomous agents | RAG & retrieval over enterprise document graphs | evaluation design for low- veri ability domains | golden-set curation & LLM-as-judge scoring | end-to-end veri able citations | iterative prompt optimization (GEPA, DSPy) | supervised ne-tuning ofmodels & embeddings (working familiarity - NeMo AutoModel, Unsloth) | multimodal document understanding (ColPali, vision models)|MCP & agent tool use Stack: Python, FastAPI, TypeScript, Cloud are Workers, PostgreSQL, SQLite, S3, Qdrant, Modal, RunPod, Pi Agent + Extensions, Claude Code, OpenAI Codex| frontier & open-weight models (Anthropic, Gemini, GPT-OSS, Gemma) PROFESSIONAL EXPERIENCE DOCUGAMI - SEATTLE, WA Founding Product Engineer | 04/2020 - Present Founding engineer at AI startup transforming enterprisedocument understanding through Vision andLarge Language Models. $40M+ funding from SignalFire, Bob Muglia, Grammarly. Discovery& Product Strategy | Conceived and shippedDocumentEngineer for Everyone (DEE) - a long-running autonomous agent that re- searches complex enterprise document sets and returns fullycited analyses - cutting customer analysis time by 80%. Pitched directly tothe CEO, secured resourcing, and led development through launch | Identi ed Commercial Insurance as the primary PMF vertical: ran outbound against the top 100 US commercial insurance brokers, narrowed to the top 20, and closed 4 with Life AD&D and Property & Casualty value propo- sitions. Opened a second vertical at a top-10 pharma company - Clinical Trial Protocol and Informed Consent Form analysis Evaluation in Low-Veri abilityDomains | Builtevaluation from scratch where correctness demands expert judgment and no labeled ground truth exists: hand-curateddocument golden sets and hundreds of question-answer pairs, with pass/fail rubrics thatdemand multi-hop agentic work owsand penalize plausible "false-positive" evidence themodel must decline to cite - benchmarking frontier and open-sourcemodel families across real-estate, healthcare, pharmaceutical,nancial, and telecomdocuments|Enforced veri able citations end-to-end: citation golden sets over PDF , DOCX, PPTX, and XLSX corpora - an-swers citable to the paragraph,gure, chart, table, and cell - with LLM judges auto-scoring answer and citation quality, turningmodel output into evidence a regulated customer can act on | Shipped automated eval reporting- precision, recall, andsemantic overlap tracked per customer over time - then evolved the harness to crawl public records for problem-matchedcontent andauto-generate new question- answer pairs andrubrics | Closed the eval-to-improvement loop commercially: iterative prompt optimization(e.g.,GEPA) - afrontier modelreviews context, generates candidate prompts, and the harnessevolves them against eval signal -took real- estate ntech TheGuarantors from 63% to 94% labelingprecision on complex rent rolls, while seven eval rounds for real-estate giantJLLshowed private ne-tuned modelsconverging from 28% to 87% on customer data Deployedat the World's Largest Customers |Builtand deployed an o ine autonomous compliance agent for the world's most valuable consumer-electronicscompany - cross-checking productspeci cations against global telecom carrier (Verizon, DoCoMo) and 3GPP requirements(band combinations, E911,High-Speed Rail) and producinggap analyses with rationale. No frontier- model access permitted: ran entirely o ineon the customer'sown infrastructure|Synthesizeda litigation-records customer's three legacy systems into a uni ed integrationarchitectureaftera single discoverymeeting;the published workow became theirproduction blueprint.Customer now processes 4M documents/month |Deliveredacomplete P&C insurance Schedule ofValues and Loss Run analysis in 4 hours - ingested, parsed, structured, and returnedin the prospect'sspeci ed JSON format. Converted to a payingcustomer | Extended the coreplatform judiciously rather thanforking itper customer- balancing ingestion, document- graph, andAPI investment against customer-speci c transforms and connectors(Salesforce,Box, Dropbox, SharePoint, SmartSheet,Airtable, Coda) Revenue & Impact | Directly responsible for 78% of rst $3MinARRthrough2025,growing to 15+enterprise clientsacross insur-ance, pharma, technology, and litigation records - including6-gure deals withtrillion-dollartechnology rms.Owned ARR, NRR expansion, and churn, and drove enterprise deal cycles end-to-end - from prospect-specicdemosto PoC scoping to dollar terms and SLA commitments | Product lead for Document Ingestion Pipeline, Table Parsing,andAuthentication/API architecture-processing 10M+ pages monthly with industry-leading bounding-box precision. The company's mostcross-functional con- tributor across engineering, product, sales, and customer success IMAGEWORK TECHNOLOGIES - NEW YORK,NYHead of SolutionsEngineering | 2001- 2019 Solutions Engineer Senior Solutions Engineer Solutions Architect Head of Solutions Engineering Ledatechnology productsrm serving state and local government agencies and global nancial institutions. Regularly embeddedat clientsites for discovery,architecture reviews, and go-lives. Grew from services to 85% subscription rev- enue in veyears.|ArchitectedClaimConnect -aRevenue Cycle Management and claims processing system handling $2.1B an- nually for NYC uniformed workers(NYPD, FDNY),NY StateWorkers'Compensation Board (WCB), and NY State Insurance Fund (NYSIF). Deployed into city and state agency infrastructure with full UAT cycles alongside agency personnel | Built ClaimConnect's medical coding and reimbursement engines-ICD-to-CPT cross-walks, DRG grouping, and NCCI-compliant code edits - reducing claim denials and accelerating adjudicationfor WCB andNYSIF | Helpedbuild Sumitomo Mitsui Financial Group's (SMFG) core Document Management System across their New York,San Francisco,and Londono ces - applying OCR and AI-driven extraction to power Know Your Customer (KYC) and Anti-Money Laundering(AML) compliance workowsfor aglobal banking institution (2003-2016) | Built and manageda 65-person distributed engineering teamacross 4 countries;additionalmajor deployments for Cartier and Walgreens Stack: Java, Spring Framework, .NET (C#), PostgreSQL, Oracle, SQL Server, Coherence Cache, Hibernate, iBATIS, JSF , JSP , Akka Actors ENTREPRENEURIAL VENTURES Founder and Technical Lead|2009 - Present | Murch.ai(2026):Agentic platform that builds collectibles dossiers from photos - multi-step identi cation, valua-tion, and cataloging.Inbeta at murch.ai | Barkie Labs (2025):Co-founder building multimodal AI for golf - ScoreShot (vision-basedscorecard capture)and Broadie (real-time voice journaling). iOS app in TestFlight beta. Live at barkie.ai | Quipio (2013-2018):Social platformreaching 2.5M users and 1.1M posts. Featured twice by Apple's App Store | CiviGuard (2009-2019):Emergency communication platform; built NYC's Advance Warning System - 65,000+ residents served during hurricanes, still in production today. Named among America's 100 Most Brilliant Companies by Entrepreneur Magazine|Alma Safety (2014-2016):911 location-accuracy hardware/software; ndings in uenced FCC requirements Additional Stacks:Erlang, Redis, Mnesia, Python/Cython (CiviGuard) EDUCATIONMIT Sloan School of Management- Cambridge, MA MBA in Innovation andGlobal Leadership (2020)| GPA:4.5/5.0 | Sloan Fellow, Enterprise Management Track | MIT-NASA BIGIdeas Challenge Winner ($163,000 Award) | Enterprise project: Second LifeBatterynancial models, General MotorsSingularity University- Mountain View, CAGraduateStudies Program - AI and Robotics Track(2009) Pace University- Pleasantville, NY BBA Information Systems (2001)| GPA:4.0/4.0in Major | Completed degree in 2 years PATENTS, PUBLICATIONS, CERTIFICATIONS & MENTORSHIP | Co-inventor on 6 granted US patentsin document intelligence, semantic annotation, and cross-document pro- cessing (Docugami, Inc., 2022-2024) | Mentor: MITSloanGenAI Lab and MIT AI Venture Studio (2025)|MIT MediaLab Diploma: AI-Generated Media (2021) | Stanford CS186C Blitzscaling,ReidHoman (2015) | Author, Pro JSF2 and HTML5, Apress(2013); Contributing Author,97 Things Every Software Architect ShouldKnow,O'Reilly (2009) OTHER Ultra-Marathon Open WaterSwimmer(English Channel) | Black-belt Budokan KarateZUBIN RUSTOM WADIA AppliedAI |Forward-Deployed Engineering(914) 498-9929 | zwadia@gmail.com | Chappaqua, NY linkedin.com/in/zubinwadia | murch.ai |barkie.aiPROFESSIONAL SUMMARY Forward-deployed productengineer who builds AI products the hard way: deep customer discovery to ndthe realprob- lem, rigorous evaluationtode necorrectness- even in low-veri ability domains where there is no ground truth-and judicious extension of thecore platform tolandthe world's largest customers without fragmenting it. Directly responsiblefor78% of therst $3M in ARR at an AI document-intelligencestartup; earlier architected a $2.1B/year government claims platform.APPLIED AI&ENGINEERING Applied AI:long-runningautonomousagents | RAG & retrieval over enterprise document graphs | evaluationdesignforlow- veri ability domains | golden-setcuration& LLM-as-judge scoring | end-to-end veri able citations | iterative promptoptimization (GEPA, DSPy) |supervised ne-tuningofmodels & embeddings (working familiarity - NeMo AutoModel, Unsloth) |multimodaldocumentunderstanding (ColPali, visionmodels) |MCP & agent tool use Stack: Python, FastAPI,TypeScript, Cloud are Workers, PostgreSQL, SQLite, S3, Qdrant, Modal,RunPod, Pi Agent + Extensions, Claude Code, OpenAI Codex | frontier & open-weight models(Anthropic, Gemini,GPT-OSS, Gemma) PROFESSIONAL EXPERIENCE DOCUGAMI - SEATTLE, WA Founding Product Engineer | 04/2020 - PresentFounding engineerat AI startup transformingenterprise document understanding through Vision and Large Language Models.$40M+ funding from SignalFire, Bob Muglia, Grammarly. Discovery&Product Strategy | Conceived and shipped Document Engineer for Everyone(DEE) - along-running autonomous agentthat re-searchescomplex enterprise document sets and returns fully cited analyses - cutting customer analysis time by 80%. Pitched directlyto the CEO, secured resourcing, and led development through launch | Identi edCommercialInsurance as the primaryPMF vertical: ranoutbound against the top 100 US commercial insurance brokers, narrowed tothetop 20, and closed4 with Life AD&D and Property& Casualtyvalue propo- sitions. Opened a second vertical at a top-10 pharmacompany - Clinical Trial Protocol andInformed ConsentFormanalysis Evaluation in Low-Veri ability Domains | Built evaluation from scratch where correctness demandsexpert judgment and nolabeled ground truth exists: hand-curated document golden sets and hundreds of question-answerpairs,withpass/fail rubrics thatdemandmulti-hop agentic work ows and penalize plausible "false-positive" evidence the model must declineto cite - benchmarkingfrontierand open-source model familiesacross real-estate, healthcare,pharmaceutical, nancial, and telecom documents | Enforcedveri ablecitations end-to-end: citation golden sets over PDF , DOCX,PPTX, and XLSX corpora - an- swers citable tothe paragraph, gure, chart, table,and cell - with LLM judges auto-scoring answer andcitation quality, turning model output into evidence aregulatedcustomer can act on | Shipped automated eval reporting - precision, recall, and semantic overlap tracked per customerovertime - thenevolvedtheharness to crawl public records for problem-matched content and auto-generate new question- answerpairsand rubrics |Closed the eval-to-improvement loop commercially: iterativeprompt optimization(e.g.,GEPA) - a frontiermodelreviews context, generates candidate prompts, and the harness evolves themagainst eval signal - took real- est
Docugami — founding
product manager. Teaching machines to read business documents properly: vision +
language models, verifiable citations, six patents along the way.
ImageWork Technologies
— head of solutions engineering. Built
ClaimConnect, a claims
platform moving $2.1B a year for NYC’s uniformed workers.
maximally-entangled
# person
name Zubin Wadia
role Founding Product Manager, Docugami (2020-present)
location Chappaqua, NY, USA
contact zwadia@gmail.com
linkedin https://linkedin.com/in/zubinwadia
# current-site
path /
title Maximally Entangled
description Zubin Wadia - engineering, product and design. 100+ products for startups, F500s and the Public Sector; writing since 2008.
▌
