Vaibhav Vimil

Senior Software Engineer · AT&T

I build data platforms, and the AI agents that run on them.

I design data systems end to end, from ingestion and lakehouse storage to governance, then build the AI layer on top: agents and natural-language tools that let people ask questions instead of writing SQL. My work spans Snowflake, Databricks, Spark, and LangGraph, running in production at national scale.

Dallas, TXSnowflake · Databricks · SparkLangChain / LangGraphAzure · AWS
$1.3M

saved per year by an AI agent that resolves service tickets on its own

300GB+

datasets ingested via Snowpipe and Databricks connectors

92%

R² on Net Adds time-series forecasting

29→54%

fiber penetration lift across the Greater Lakes region

2022 – present

Experience

Four years at AT&T, from analytics and fraud ML to owning the data platform and the agents on top of it.

May 2024 – Present

Senior Software Engineer

AT&T · Automated Platforms Development · Plano, TX

  • Designed and shipped the INLAP AI Platform, a natural-language console over AT&T's Snowflake address dataset: users query by address or GLID and get a record view with DA/PSA boundary maps (MapLibre), Street View, and live serviceability metrics.
  • Built a production agent chat layer with LangChain/LangGraph backed by a self-hosted small language model on Databricks that creates and resolves service tickets with no manual intervention, saving $1.3M per year. Inference and agent orchestration stay inside the governed platform boundary.
  • Own and architect AT&T's Snowflake data platform behind nationwide U.S. address and serviceability pipelines that feed fiber build-out and network expansion planning.
  • Reduced ETL runtime from 36 hours to under 20 minutes (97%) by re-architecting VM-based JavaScript workloads onto Databricks around partitioned distributed processing.
  • Built ingestion for 300GB+ datasets via Snowpipe and Databricks–Snowflake connectors, with dbt models and Snowpark logic for anomaly detection, missing-data analysis, and national-scale address classification.
  • Enforced governance across business units with Snowflake RBAC, column-level security, masking, and row access policies, the access controls that AI tools over enterprise data must respect.
  • Leading a production migration from Snowflake native tables to Azure-backed Iceberg tables, moving enterprise storage to an open table format while validating ELT parity.
  • Scoped ambiguous business requirements into shipped tools, and delivered architecture readouts and Streamlit dashboards translating platform tradeoffs into business value for executive stakeholders.
Jul 2023 – Apr 2024

Senior Specialist, Technology Development Engineer

AT&T · Fraud ML · Plano, TX

  • Owned and deployed the production automotive fraud detection pipeline on Databricks and PySpark end to end. Credit applications are vectorized with DeBERTa embeddings and matched with FAISS k-nearest-neighbor search to flag high-similarity applications at enterprise scale, cutting processing time 72% versus the legacy system despite messy, incomplete records.
  • Won an internal AT&T hackathon by building a low-latency connected-car telemetry pipeline integrating Palantir, Databricks, and Power BI for live anomaly detection, demoed to senior leadership.
Jul 2022 – Jul 2023

Specialist, Technology Development Engineer

AT&T · Advanced Analytics & Automation · Plano, TX

  • Replaced manual executive reporting by building Python ETL pipelines that integrate Excel, Palantir, and enterprise data into Snowflake, cutting pipeline execution time 83%.
  • Built time-series forecasting for Net Adds (92% R²) and regression models identifying fiber expansion opportunities; insights supported a penetration increase from 29% to 54% across the Greater Lakes region and informed capital allocation.

Part-time · Remote · Aug 2026 – Present

AI Model Evaluator, Mercor

Two part-time contracts evaluating frontier AI models

  • Coding agents: evaluate how frontier AI coding agents handle software engineering tasks, judging correctness and code quality and writing structured feedback.
  • Text and reasoning: grade model responses on general text and reasoning tasks against detailed rubrics, with written feedback used to improve model quality.

At AT&T

Selected work

Simplified diagrams of systems running in production. The moving lines show which way data flows.

AI platform

INLAP AI Platform

A natural-language console over the Snowflake address dataset. Users look up an address or GLID and get a record view with DA/PSA boundary mapping, Street View, and live serviceability metrics.

300GB+datasets flowing in through Snowpipe and Databricks
INSIDE THE PLATFORM BOUNDARY Your questionaddress, GLID,or plain language Agent layerLangGraph agentsself-hosted SLM Snowflake platformRBAC · maskingrow access policies Record viewDA/PSA boundariesStreet View · metrics IngestionSnowpipe · Databricks300GB+ datasets Service ticketscreated and resolvedno manual hand-off Your questionaddress, GLID, or plain language INSIDE THE PLATFORM BOUNDARY Agent layerLangGraph · self-hosted SLM Snowflake platformRBAC · masking · row access Ticketsauto-resolved Ingestion300GB+ data Record viewDA/PSA · Street View · metrics

Simplified flow. Inference and orchestration run inside the governed platform, so data never leaves it.

SnowflakeMapLibreNatural-language query
Agents

Agent chat layer

LangChain/LangGraph agents on a self-hosted small language model on Databricks that create and resolve service tickets on their own, with no manual hand-off. Inference and orchestration stay inside the governed platform boundary.

Issue comes inuser request or alert Agent decidesLangGraph · self-hosted SLM Creates the ticketlogged automatically Resolves and closes itno manual hand-off Snowflakerecords
$1.3Msaved per year by automating service tickets end to end
LangGraphSelf-hosted SLMDatabricks
Lakehouse

Iceberg migration

Moving enterprise storage from Snowflake native tables to Azure-backed Apache Iceberg tables, an open table format, while validating that ELT output stays identical.

Before: Snowflake native tablesstorage managed inside Snowflake MIGRATE After: Apache Iceberg tablesopen table format on Azure storage Snowflakequeries as before Other enginese.g. Spark, same data ✓ ELT OUTPUTS VALIDATED TO MATCH
In flightleading the production migration now
Apache IcebergAzureELT parity
ML · Fraud

Fraud similarity search

Each credit application becomes a DeBERTa embedding, a point in vector space. FAISS finds its nearest neighbors, and applications that land almost on top of each other get flagged as likely fraud. The production pipeline cut processing time 72% versus the legacy system.

−72%processing time versus the legacy system

Illustration of the approach, not real data.

DeBERTaFAISS k-NNPySparkDatabricks

Nights & weekends

Building now

Independent projects, from a property marketplace to a robot dog.

Building

Homes IQ Founder

An owner-to-owner property marketplace where buyers and sellers close a deal directly, without a broker. LLM agents generate, populate, and review the transaction paperwork, with evals and human review on the fields that matter. Ingestion pipelines normalize, deduplicate, and incrementally refresh listings and property records on a low fixed-cost budget.

Next.jsPythonLLM agentsCloudflare R2
Running

Hermes trading agent

An always-on personal agent hosted on AWS that tracks and trades my Robinhood portfolio through an MCP connection. Email-report and general-assistant agents are next.

AWSMCPRobinhood
Running

Prediction-market bot

An LLM-driven bot that trades event contracts on Kalshi through the Kalshi API.

Kalshi APILLMPython
In development

Rent an Agent

A marketplace of specialized AI agents for people who would never build their own. It starts with a bills agent that connects bank accounts through Plaid and handles bills end to end, including paying them.

Next.jsTypeScriptPythonPlaid
In development

Shorts generator

A personal tool that produces YouTube Shorts from fully AI-generated cinematic clips instead of stock footage, running on Azure.

AzureVideo generation
In development

Football draft game

A daily dream-team draft game for American football: pick a lineup each day and see how it stacks up.

MobileGame design
Exploring

Robot dog

A 3D-printed quadruped. I'm working out the motors, components, and the AI that will drive it.

3D printingRoboticsEmbedded AI

Earlier work, including my UT Dallas team project, is on GitHub ↗︎.

Toolkit

Skills & languages

Data platforms

SnowflakeSnowpipeSnowparkIceberg tablesDatabricksDelta Lakedbt

AI & ML

LangChainLangGraphLLMsTransformersEmbeddings & FAISSForecastingAnomaly detection

Processing

Apache SparkPySparkDistributed systemsSQL optimizationELT

Governance

RBACColumn-level securityData maskingRow access policiesPII handling

Cloud & infra

Azure ADLSData FactoryAzure DatabricksAWS S3GlueAthenaRedshiftEMRSageMakerCloudflare R2DockerKubernetesGit

Languages

PythonSQLPySparkBash

Education & certs

Credentials

Degree

B.S. Computer Science

The University of Texas at Dallas

2018 – 2022

Patent

V2X networking

Filed

Current

SnowPro Core Certified SnowflakeValid to Jan 2027
AWS Certified Solutions Architect – Associate AWSValid to Jan 2028
Generative AI with Large Language Models DeepLearning.AIDec 2023
Databricks Certified Data Engineer Professional DatabricksIn progress

Earlier

Databricks Lakehouse Fundamentals DatabricksEarned 2023
Certified SAFe 5 Agilist Scaled AgileEarned 2023
Azure AI Fundamentals MicrosoftEarned 2022
Azure Data Fundamentals MicrosoftEarned 2022
AWS Certified Cloud Practitioner AWSEarned 2020

Contact

Open to forward deployed, applied AI, and data platform roles.