Arun ChukkalaRésumé
Navigation
View résumé

San Francisco, CA

AI Implementation Engineer

AI that fits the way
business works.

I’m Arun. I turn AI into useful, dependable tools for the people doing the work. My focus is retail agents and connected workflows: less re-entry, fewer loose ends, and more room to run the business.

Explore my workGet in touch
Currently
AI implementation at IBM
Building
Cork · Retail operations
Based in
San Francisco, CA

01Selected work

Built around a real working day.

A closer look at Cork, and a few other ways I put AI to work.

Retail agents & operations

Cork.

More time for the store.
Less time between systems.

Independent retailers have enough to keep track of. I’m building Cork to connect checkout, inventory, purchasing, and invoices, with retail agents that handle the repetitive steps and leave the important decisions with the store team.

  • Less re-entry. Extract invoice lines and match them to the store’s catalog, with corrections remembered for the next delivery.

  • Clearer decisions. Surface uncertain matches for review before anything changes in inventory or the books.

  • Careful by design. Keep store data separate, record changes, test matching behavior, and place limits on AI usage.

Explore CorkMy focus / Product & engineering
cork.Interactive example

Start with the invoice.

01 / 03
Distributor invoiceSample delivery · INV-024
Item descriptionQuantity
VLY RED 750ML12 bottles
CST WHT 750ML6 bottles
RSV RED 7506 bottles
3 lines ready for matching

Extract the lines, quantities, and prices from a distributor’s document. Give the team structured information to work with.

Illustrative data. No live AI call or store records are changed.

Built withTypeScriptReactNode.jsPostgreSQLPrismapgvectorGemini
Watch Cork demo

Silent product walkthrough. Open video

Other implementations

Different problems. The same care.
AI Medical-Bill Auditor02

Claim Guardian

Turns a medical bill into a review of its charges, using CMS rate references and cited explanations. Calculations stay in code; the model helps make the findings understandable.

FastAPIGeminiRAGNext.js
Watch Claim Guardian demo

Silent product walkthrough. Open video

AI Job Search · LLM Evals03

Agentception

Matches a résumé against live roles and explains where the fit is strong or incomplete. An evaluation harness checks the ranking against real job descriptions.

FastAPIReactVoyageSupabase
Watch Agentception demo

Silent product walkthrough. Open video

Multi-Agent Code Verification04

Code Tribunal

Checks code changes against the original requirements, using multiple agents to surface omissions and scope changes. Available through a web interface, CLI, and MCP server.

PythonFastAPIMCPNext.js
Watch Code Tribunal demo

Silent product walkthrough. Open video

Compliance-Grade Financial RAG05

FinAgent

A self-hosted research workflow over SEC filings. Combines retrieval, source validation, and an audit trail so teams can inspect how an answer was produced.

PythonLangGraphQdrantNext.js
Watch FinAgent demo

Silent product walkthrough. Open video

02How I work

From a useful idea to a useful system.

The implementation matters as much as the model. These are the parts I pay attention to.

01

Start with the workflow

Find the repeated handoffs, manual checks, and missing context. Define what a useful result looks like before choosing a model.

  • Workflow mapping
  • Document intelligence / OCR
  • Human-in-the-loop design
  • Evaluation datasets
02

Connect the right pieces

Bring models, business data, and existing tools together. Give agents a defined job and clear boundaries for what they can change.

  • Agent workflows / LangGraph
  • Retrieval & tool calling
  • Model Context Protocol (MCP)
  • Python / TypeScript / React
03

Make it dependable

Test with real examples, make failures visible, and keep usage costs in view. Build something a team can operate after the demo.

  • Regression gates & guardrails
  • Tracing & cost controls
  • FastAPI / PostgreSQL / Prisma
  • Docker / CI/CD / cloud deployment

The goal: less operational friction, with a clear view of quality, cost, and control.

03Experience

A foundation in shipping and maintaining.

Enterprise engineering experience, carried into hands-on product building.

IBM

Mar 2024 — PresentUnited States

AI implementation & agent systems

  • Design and maintain evaluation infrastructure for LangGraph-based agent workflows in a regulated enterprise environment — golden datasets, regression gates, human-in-the-loop review — reducing manual review time by 30%.
  • Reduced agent workflow latency by 62% (850ms → 320ms) through asynchronous execution and semantic caching of LLM calls.
  • Ship Dockerized Python/FastAPI services orchestrating LLMs, internal tools, and enterprise data through CI/CD — 98%+ uptime with retry, timeout, and fallback handling around non-deterministic model calls.
  • Build production observability that traces agent execution, tool calls, and failure modes end to end, converting production failures into permanent regression tests.

Infosys

Jan 2021 — Dec 2022Hyderabad, India

Software Engineer, Applied AI / Risk Platform

  • Engineered ensemble anomaly-detection workflows for fraud decisioning, reducing false positives by 18% on a regulated risk platform.
  • Designed and deployed Dockerized FastAPI microservices and Azure Data Factory ETL pipelines processing 15M+ records.
  • Implemented output-consistency monitoring that alerted on model drift before it reached downstream decisions.

Education

M.S. Computer ScienceLamar UniversityDec 2024
B.Tech. Computer ScienceSri Indu Institute of Engineering & TechnologyMay 2020

A note on agent safety

Preprint — 2026 / Agentic AI Safety

BARC: Blast-Radius-Aware Agentic Control

Agentic systems should be controlled by the consequence of their actions — not by model confidence. The paper introduces a tiered safety model for tools: read-only, draft, bounded write, and irreversible.

Read the paper (PDF)