HARISH
full-stack systems, llm orchestration, production-minded engineering.
i'm harish.
Engineer focused on AI systems, full stack delivery, and LLM workflows that behave when the traffic is real. Finishing my degree at UT Knoxville in May 2026 (Information Sciences, AI & Data concentration).
Four projects shipped end-to-end: a bounded-agent AI system, a policy-routed internal platform, an LLM-driven QA loop, and a live-data job tracker. Each one taught me a different lesson about shipping under real constraints.
the usual suspects.
the stack, grouped how i actually reach for it: cloud, frontend, backend, AI and quality.
the work.
Four real builds. Each one deployed, each one taught me something I couldn't learn from tutorials.
Evident
Evidence-grounded AI decision system
Most LLM decision systems hallucinate confidence. Evident ranks research outreach targets but refuses to answer when the evidence is too thin.
- Hybrid pipeline: deterministic extraction → enrichment → prefilter → selective Claude evaluation. Cuts unnecessary model calls by ~60%.
- Bounded agent loop: evaluation → adaptive retrieval → second-pass review → final decision, capped at one retrieval per run.
- Productionized on AWS ECS / Fargate with Docker and Secrets Manager. Demo and local modes with per-run cost ceilings.
Confident rankings with explicit refusals when evidence is weak, full citation trails, and predictable cost control that stays stable under real usage.
StackGate
AI-assisted internal developer platform
Engineers ask for infra in Slack. Tickets get lost. Approvals are messy. StackGate turns plain-English requests into validated, policy-routed tickets.
- Risk and policy engine classifying requests into low / medium / high tiers, auto-approving safe ones, routing the rest through a role-aware approval chain.
- A modular provisioning layer that lets providers plug in cleanly while the core workflow stays stable, with guarded Azure Postgres paths for low-risk dev requests.
- Complete audit trail capturing every workflow event with full requester-facing transparency on routing decisions.
Infra requests go from chat noise to traceable, policy-aware tickets, with humans still in the loop where it matters.
VeriFlow
AI-powered QA automation copilot
User stories live in one tool. Tests live in another. The gap between them is where coverage dies. VeriFlow closes the loop.
- Converts user stories into executable Playwright tests, runs them in a real browser, and persists the run history.
- Integrated Claude with Playwright to validate live web app workflows end-to-end.
- Pulls user stories and work items directly from Azure DevOps APIs as input to test generation.
Written requirements become automated coverage. Scheduled reruns, persisted history, fewer regressions making it past CI.
ApplyTrack
Job application tracker with live job search
Job hunt spreadsheets break the moment you start applying seriously. ApplyTrack turns the chaos into a real workflow with live job listings baked in.
- Live job search via Adzuna pulled through a server-side proxy — no API keys exposed in the browser.
- Workflow tracker (saved → applied → interview → offer / rejected / withdrawn / archived) with edit-in-place that doesn't lose state.
- Real auth via Supabase, hosted Postgres, and a demo account so you can poke around without signing up.
A focused, no-fluff job tracker that's actually pleasant to live in for months at a time.
Evident is stepping out of the lab.
This rollout is built to capture real user behavior and real hiring signal from resumes, then turn that feedback into hard performance wins and measurable lift.
let's talk.
Graduating May 2026. Open to internships and new-grad roles in AI systems, infra, or full-stack. Nashville-first, remote-friendly, will relocate for the right team.
