Applied AI engineer — Austin, Texas

AI systems built for real business workflows.

I build production AI systems and the data pipelines behind them. At Welltower, I take workflows from business requirements through deployment and adoption, drawing on nine years across enterprise software, data engineering, and applied AI.

See selected work Download the resume

Senior AI Analyst, Welltower Inc.
MS Computer Science, Seattle University.

~1 FTE

recovered annually by automating payroll tie-out

~$100K

use tax penalty exposure resolved

most-used internal GPT company-wide, six months running

departments running systems I shipped

Governing design principle

Keep calculations deterministic. Make AI outputs reviewable.

I combine model-based extraction and recommendations with source evidence, validation, and human review. Reconciliation workflows propose ledger changes for approval; the HR assistant checks location before returning regional policy guidance.

Selected work — Welltower Inc.

From business process to production AI.

Selected systems I own from discovery and architecture through user acceptance testing, deployment, monitoring, and iteration. Open a project to explore the engineering decisions and outcomes.

Shared workflow orchestration platform

Welltower · n8n · reusable sub-workflows · monitoring

A shared foundation for repeatable department automations, with retry and error recovery built into the workflow architecture.

departments served by production AI and automation

Read project details Close details

Problem. Department automations needed a repeatable path from one-off experiments to supported production workflows.
My contribution. Architected a queue orchestrator and reusable sub-workflow library with atomic retry and centralized error recovery. New department automations can use the shared foundation without changing the orchestrator.
Delivery. Documented 25–30 processes across 5–6 departments and owned requirements, build, user acceptance testing, deployment, logging, and failure monitoring.

Webb — enterprise HR assistant

Human Resources · ChatGPT custom GPT · OpenAI Compliance API · Snowflake

An HR assistant with regional policy routing, regression checks, and a review pipeline for production conversations.

85%

fewer HR support tickets

Read project details Close details

Adoption. Migrated the company-wide assistant from a legacy Slack bot to a ChatGPT custom GPT. It reached 700 users, 200 monthly active users, and 100+ daily queries, becoming the most-used internal GPT for six months.
Controls. Added location-based routing, limits on extrapolation from partial documents, and prompt-injection defenses. Gated revisions on regression cases covering policy answers, regional routing, refusals, and adversarial prompts, plus a 10-day Human Capital review.
Production feedback. Built an OpenAI Compliance API pipeline landing conversations in Snowflake, with a review frontend and human flagging. HR uses the resulting review process weekly.

Journal entry tie-out automation

Accounting · n8n · deterministic reconciliation

Payroll cash reconciliation, journal entry preparation, and suspense clearing, with proposed ledger changes reviewed by a person.

~1 FTE

of accounting capacity recovered

Read project details Close details

Before and after. Payroll tie-out required 7–8 accounting team members, each spending 3–4 hours weekly per operator. The automated workflow brings the work across all operators to approximately four hours per week, recovering about one FTE of capacity.
My contribution. Implemented payroll cash tie-out, journal entry preparation, and suspense item clearing as production n8n workflows. Calculations and reconciliation stay in deterministic code.
Approval boundary. The workflow proposes missing chart-of-accounts entries and mapping corrections for human approval. It does not apply those changes to the general ledger.

Use tax compliance automation

Tax · n8n · Thomson Reuters tax code data

Use tax determination for state, city, and county codes across 3,000+ invoices monthly and 1,500+ properties.

~$100K

penalty exposure resolved

Read the Project Details Close

Resolved approximately $100K in use tax penalty exposure and established the standing control preventing recurrence across all operating jurisdictions — automating use tax determination for state, city, and county codes for 3,000+ invoices monthly across 1,500+ properties.
Eliminated a hallucination risk unacceptable in a compliance calculation — rebuilt an Azure live-search-per-line-item design as an n8n workflow grounded in Thomson Reuters tax code data: a reference-data lookup handles known commodities, unmatched lines escalate to a higher-capability model returning a cited proposal, and line-total mismatches route to human review.

Insurance requirements extraction

Legal · Amazon Bedrock · document intelligence

Extracts insurance requirements from acquisition agreements and compares coverage limits with historical deal benchmarks.

Cited

limits linked to source clauses and pages

Read project details Close details

My contribution. Built a three-stage Amazon Bedrock workflow: identify deal context, extract clauses with verbatim quotes and page citations, then normalize free-text limits to standard coverage types.
Design decision. Kept benchmarking in deterministic code. A separate engine compares each extracted limit against historical deal statistics and flags it for review, preserving the source evidence alongside the result.

Construction scope generation

Capital Projects · guided generation · Word documents

Turns project managers’ inputs into scope-of-work documents while preserving the formatting of the source templates.

trade categories in the scope library

Read project details Close details

Grounding. Built guided generation around 1,022 scope items, 16 templates, and a 2,384-property dataset. A climate and wind pass adapts specifications to the building’s environment.
Engineering detail. Used a similarity-based merge to align generated text with original Word paragraphs, preserving numbering, styles, and run-level formatting. Low-confidence matches retain the original text.
Adoption. The tool replaced manual per-project drafting and is used by every project manager on the team.

Portfolio analytics in natural language

Investment · Snowflake · natural language to SQL · Claude plugin

Lets the Investment team query property data and produce reports directly in chat.

1,500+

properties available for analysis

Read project details Close details

My contribution. Built a natural-language-to-SQL system that gives the Investment team access to portfolio analysis without relying on the data team for each ad-hoc request.
Iteration. Replaced the initial Streamlit interface with a Claude plugin pairing a Snowflake connector with skills that encode the team’s output formats and report structures.

Property rent recommendations

Asset Management · custom GPT · Excel workbooks

Generates site maps and experimental rental amount recommendations for asset management to review.

Human review

rental changes approved, rejected, or adjusted

Read project details Close details

My contribution. Created the Excel workbook format and a custom GPT that generates site maps and experimental rental amount changes. The workflow accounts for unit location, area, proximity to amenities, and available services.
Decision boundary. Asset management reviews the recommendations and approves, rejects, or adjusts the proposed changes before adoption.
Outcome. A comparison between two buildings at one property showed improved rental income and overall profit with the new approach. The workbook format was extended to additional properties.

Call center analytics

Operations · transcript analysis · Power BI

Moves call analysis from third-party batches to daily internal review, connecting conversations with conversion and occupancy.

15%

additional leads recovered from dropped conversations

Read project details Close details

My contribution. Replaced a third-party transcript analysis service with an in-house AI system for daily review of 3,000+ calls across 50+ properties. Built Power BI reporting that connects call topics to occupancy and conversion.
Outcome. Recovered 15% additional leads from dropped conversations and contributed to a 3% revenue increase. Property managers adopted the dashboards as a standing KPI review.

Enterprise delivery — Yardi Systems

Integrations that teams can run and extend.

Before building enterprise AI, I delivered SaaS integrations and BI implementations with client teams, from requirements and data validation to rollout and training.

Lease certification integration

Dubai · Python · SQL · REST APIs · transaction monitoring

Connected Yardi’s lease workflow to the external Ejari certification service, replacing a manual submission process.

Minutes

certification turnaround, previously about a week

Read project details Close details

My contribution. Led the integration design and implementation with the certification service’s engineers. Mapped lease data from SQL into staging tables, transformed it into API payloads, and added validation, error handling, and transaction logging.
Rollout. Coordinated client stakeholders and engineering teams through testing and deployment. Signed agreements could be transmitted automatically instead of being printed, scanned, and submitted manually.
Scale. Adoption grew from five clients in year one to 30–35 by year three, before the integration became a standard offering.

Reusable BI implementation package

Dubai · SQL Server · data warehousing · Orion BI

Consolidated years of client-specific BI work into a shared package that engineers could deploy and clients could extend.

~120

KPIs consolidated into one deployment package

Read project details Close details

Problem. Client-by-client copies of KPI logic created redundant calculations, inconsistent definitions, and repeated implementation work.
My contribution. Built a consolidated package from six to seven years of accumulated requirements. Deduplicated KPI logic, dashboards, and measure sets, then standardized deployment and client-specific configuration.
Delivery impact. Reduced implementations that previously took months to about two weeks. Trained business users on reporting, administrators on access and dashboard setup, and client engineers on adding measures and custom SQL.

Leasing implementation recovery

Pune · workflow configuration · requirements discovery · adoption

Reworked a stalled client implementation by examining how users actually completed their daily tasks.

$3M

contract retained: $1M annually over three years

Read project details Close details

Discovery. Joined an engagement after two unsuccessful implementations and worked directly with users on-site to understand reporting, leasing, and document-generation gaps.
My contribution. Rebuilt configuration, approval workflows, automations, reports, and proposal templates. Delivered individual training and demonstrated native BI capabilities to reduce reliance on manual exports.
Outcome. Monthly reported issues fell from approximately 100 to 7–8, with client sign-off within two months. The recovery helped retain the three-year contract and contributed to my promotion to the Dubai team.

Research projects — github.com/jay-singhvi

From research to a working search system.

Resonate — research to lecture search

Seattle University · LangChain · Pinecone · AWS Transcribe · Streamlit

A team capstone and published RAG research project that developed into a lecture-search system used across five academic departments.

89%

precision and recall in project evaluation

Read project details Close details

Research foundation. Co-developed Resonate as a capstone for retrieving information from recorded meetings. The work led to “A Retrieval-Augmented Framework for Meeting Insight Extraction,” published at ACM SAC 2025.
My contribution. Built the path from recording ingestion and transcription to chunking, embeddings, retrieval, and a Streamlit search interface. Following the research project’s success, developed the approach into lecture search adopted by five academic departments.
Evaluation and optimization. The project reported 89% precision and recall. Work on inference and caching reduced response latency by 35%, while Llama 2 QLoRA fine-tuning reduced compute requirements by 50%.

View on GitHub

Deployment automation at Seattle University

Built AWS CloudFormation templates to replace manual console setup across 50+ integrations, reducing deployment time by 65%. Paired the rollout with cross-team verification, documentation, and knowledge-transfer sessions.

Explore research repositories

Experience

Nine years across software, data, and AI.

Aug 2025 – Present

Welltower Inc.
Dallas, Texas

Senior AI Analyst - Applied AI Engineering & Enablement

Shipped production AI systems adopted across six departments — ChatGPT custom GPTs, Streamlit applications, and n8n workflow automation — owning each end-to-end from department requirements through UAT, deployment, logging, and failure monitoring.
Architected the shared automation platform for department workflows — a queue orchestrator with a reusable sub-workflow library, atomic retry, and centralized error recovery, so adding a new department automation requires no orchestrator change.

Sep 2022 – Aug 2025

Seattle University
Seattle, Washington

Data Scientist — research assistant

Predicted asthma onset at 88% accuracy — 20% better than traditional methods — with transfer-learning and PySpark ensemble models on sparse medical data, improving forecasting accuracy by 12% over standard neural networks, on HIPAA-compliant Docker and AWS pipelines.

Apr 2019 – Jul 2022

Yardi Systems
Dubai, United Arab Emirates

Technical Consultant - SaaS Implementation

Personally owned 20+ SaaS implementations across Commercial Leasing Manager and Orion BI over my time at Yardi, including approximately 15–16 BI implementations in Dubai. Built the reusable KPI package that standardized BI delivery.
Delivered requirements discovery, data modeling, SQL development, user acceptance testing, executive demos, and go-live support. Restored trust in a BI engagement by tracing KPIs to source data, securing report-level sign-off, and co-presenting with client managers.

Nov 2016 – Mar 2019

Yardi Systems
Pune, India

Software Engineer

Developed and configured leasing workflows, reports, and automations for enterprise clients. Recovered a stalled implementation through on-site discovery and training, helping retain a $3M contract over three years.
Powered real-time dashboards across 100+ KPIs with a dimensionally modeled data warehouse and data marts — 15% faster processing via incremental loading — and set the governance standards, automated testing, and query tuning behind them.

Technical skills

The stack behind it.

AI and machine learning

Python · PyTorch · RAG · LLM evaluation · embeddings · vector databases (FAISS, Pinecone) · prompt engineering · guardrail and anti-hallucination design · OpenAI (GPT, custom GPTs, Compliance API) · Anthropic Claude (plugins, skills, MCP) · Amazon Bedrock · LangChain · Hugging Face · Scikit-learn

Data and analytics

Snowflake (Cortex Analyst) · SQL · T-SQL · PostgreSQL · PySpark · Spark SQL · Pandas · NumPy · ETL pipeline development · SSIS · dimensional modeling · data warehousing · incremental loading · query optimization

Cloud and infrastructure

AWS (S3, SageMaker, Transcribe, Bedrock, CloudFormation) · Docker · Git · GitHub Actions CI/CD · RESTful APIs

Applications and reporting

Streamlit · n8n · Power BI · Tableau · KPI dashboards · Excel

Education

MS, Computer Science — data science

Seattle University, Seattle, Washington · Sept 2022 – Jun 2024

MS, Computer Applications

Symbiosis International University, India · July 2015 – Apr 2018

BS, Information Technology

University of Mumbai, Mumbai, India · Jun 2011 – Jan 2015

Research papers — /publications

Publications and certifications

DAWAK 2024

Incremental SMOTE with Control Coefficient for Classifiers in Data Starved Medical Applications

SAC 2025

A Retrieval-Augmented Framework for Meeting Insight Extraction

IEEE 2026 - Accepted for publication

A Hybrid Deep Learning Framework using Transfer Learning as the Feature Extractor in Environmental Health Risk Prediction

AWS Certification

AWS Certified AI Practitioner (AIF-C01)

Get in touch.

Open to AI Engineer, AI Analyst, Forward Deployed Engineer, and AI Data Engineer roles.

jay.singhvi.1993@gmail.com 206-660-3609

LinkedIn GitHub Google Scholar Resume