HS logo

HARISH

i build
agentic systems
that hold up in production.

production systems
built for real constraints

built for desktop. optimized for mobile.

Harish Sakthivel
Harish Sakthivel
/Python//LLM APIs//OpenAI API//AWS//React//Next.js//TypeScript//Framer//Lenis//FastAPI//Playwright//Postgres//Docker//Terraform//GitHub Actions//LLM orchestration//Supabase//Prisma//Node.js//Vercel//Streamlit//Azure//Claude Code//Power Automate//SharePoint//Jira//Bitbucket//Python//LLM APIs//OpenAI API//AWS//React//Next.js//TypeScript//Framer//Lenis//FastAPI//Playwright//Postgres//Docker//Terraform//GitHub Actions//LLM orchestration//Supabase//Prisma//Node.js//Vercel//Streamlit//Azure//Claude Code//Power Automate//SharePoint//Jira//Bitbucket/
02 / who's behind this

the engineer

I build production systems that check their own work: evidence before an answer, a rule before an action, a quote before a stored value.

In my first three months as an AI & Automation intern at Blue Diamond, I shipped four production systems: a live quoting platform reps use on calls, monitored by an eight-agent fleet; a source-grounded spec pipeline; a warehouse capture app running daily on the floor; and end-to-end HR document automation. Before that: UT Knoxville, Information Sciences (AI & Data), Summa Cum Laude.

The through-line is verification and cost: nothing gets stored or recommended without provable backing, and the model only runs where deterministic code can't.

based in
Nashville, TN
currently
AI & Automation Intern Blue Diamond Attachments
school
UT Knoxville
degree
Information Sciences (AI & Data)
gpa
3.75 / 4.0 Summa Cum Laude
open to
full-time new grad roles or a strong internship
03 / where i've worked

experience.

internships and professional work. more being added.

AI & Automation Intern

May 2026 – Present
Blue Diamond Attachments · Knoxville, TN

Sole developer of the company's internal platforms (a live sales-quoting tool with an AI assistant, a shipped warehouse desktop app, and end-to-end document automation) and lead developer of the machine-spec pipeline behind its product configurator.

  • Replaced a 76-page PDF price guide with a live quoting tool reps use on calls: 11,491 SKUs, six dealer tiers, live inventory, and 1,181 machine specs, deployed on Vercel from a Python pipeline (SyteLine ERP through Zapier into Supabase). Re-read the category hierarchy from the source guide to fix a regex bug that had mis-bucketed 810 SKUs, reaching 0 duplicates and 0 unassigned.
  • Shipped an AI assistant into that tool that turns a rep's spoken request or a full call transcript into a priced quote, running Claude in a serverless function while the browser searches the catalog locally, keeping per-query cost down and the API key off the client.
  • Built an autonomous 8-agent system that audits the quoting platform every 30 minutes against 11 machine-checkable rules (39 tests) at near-zero model cost, coordinating on a shared findings bus, filing findings as Jira tickets, and opening PRs for proven fixes while leaving the merge to a human. Caught price drift worth roughly $4,000 per unit and a missed increase that left a batch of items under-priced.
  • Led the machine-spec pipeline behind the company's product configurator: 4,447 specs across 628 of 1,314 models, each backed by a verbatim source quote and any ungrounded value nulled, enforced by one check shared by the writer and the auditor. Flagship categories at 98.6% core-field fill across 172/172 models, cross-checked against a second source; spec requests are filed as Jira tickets an agent then fulfills.
  • Cut misfiled shipping photos to zero with a self-updating Python desktop app (OpenCV, Tkinter) that auto-files roughly 1,400 photos a week (about 6,000 a month) into the correct SharePoint library across two departments, with path auto-discovery and a local fallback so a photo is never lost.
  • Built and deployed four Power Automate flows that convert Microsoft Forms submissions into formatted Word/PDF safety, injury, and performance reports, auto-stored to SharePoint and emailed out, eliminating manual formatting and routing across ~360 documents a year (an estimated 215 hours), then trained HR to build and maintain their own.
  • Launched the company's first internal AI program as its appointed spokesperson, onboarding a third of staff in the first rollout.
PythonTypeScriptClaudeVercelSupabaseOpenCVNode.jsPower AutomateSharePointStreamlit
04 / things i've shipped

the work.

production systems built under real constraints. all deployed. all in use.

Blue Diamond Attachments

Independent Projects

05 / platforms & tools

the stack.

grouped by how i actually use it

tap logos for usage context
infrastructure
frontend
backend & data
AI & quality
automation & workflow

up next.

Drive Companion

in-car AI that knows if you'll make it

Learns your car's real fuel efficiency from live OBD-II telemetry and answers "can I make it?" in real time, pairing a personalized ML model with a deterministic range engine.

Map, chat, and voice in one companion, where the LLM handles language and the math stays deterministic.

private beta · live APIs

request access
06 / let's talk

get in touch.

Currently an AI & Automation intern at Blue Diamond Attachments, and open to full-time new grad roles (or a strong internship) in AI systems and production engineering. Based in Nashville, remote-friendly, and open to relocation. Email is the fastest way to reach me.