Live in production
Rudy
2026A campus safety assistant that answers from verified documents or hands off to a human.
Campus safetySolo build
301 adversarial checksView case study
AI & data systems engineer · Cal Poly
San Luis Obispo, California
Rudra Khakhar
Cal Poly Computer Engineering, Class of 2026 · Building at Cal Poly EHS & Risk Management
Open to roles & select projectsSelected work
Ideas meet
the real world
Tools people use. Systems put to the test. A look inside the decisions, the evidence, and the edges of each build.
Sandbox · synthetic data
Shockwatch
2026An insurance renewal sandbox that blocks approval when the evidence does not add up.
Insurance renewal workflowsSolo build
114 passing testsView case study
Production · team access
OFSM Inspection Tracker
2026Inspection reports become tracked violations and the Excel exports the team already uses.
Fire and life safetySolo build
Production · team access
Fall Protection Registry
2026A scattered archive becomes one searchable registry, with a source behind every record.
Campus complianceSolo build
Live in production
The Sentinel Protocol
2026A physics-anchored, six-layer detector catches eight spoofing attacks on a destroyer, live in 3D.
Cyber-physical securityTeam project
Field research
Mesosphere
2025Adaptive power management improved mission duration and reduced power draw in field tests.
Defense telemetryTeam project
Production · team access
Enterprise Risk Platform
2026Risk scoring and document-grounded AI for a team-built Cal Poly platform.
Enterprise risk managementTeam project
Live in production
Meera Services
2026A ground-up rebuild of an enterprise AI-training company’s site, now in production.
Enterprise marketing engineeringSolo build
Work with me
Built fast.
Built to stay up
I build with Claude Code, Codex, and whichever model fits the job. They are tools that carry my speed and my creativity; the architecture, the review, and the accountability are mine. Every line is reviewed, tested, and handed to you with the repo, the docs, and the accounts.
How I build, and what you ownProof-of-Work Week
A working thing you can click by Friday.
You get
- One core flow, built end to end
- A live URL, with sign-in if the flow needs it
- The repository and a README your next engineer can follow
- A short recorded walkthrough
- A written note on what this build is not
Clock
5 business days. Staging URL by day 3.
Starts with a written scope, not a discovery call.
Ops Tool Build
Replace the spreadsheet that runs your team.
You get
- A dashboard, registry, or workflow app on Supabase and Vercel
- Sign-in, roles, and the exports your team already uses
- A data pipeline from the files you have today
- Tests on the flows that matter, plus a handoff document
- 30 days of fixes after handover
Clock
2 to 3 weeks.
Starts with a half-day look at the spreadsheet and the people who use it.
Answers-From-Your-Docs Assistant
Cites your documents. Says “I don’t know” when it should.
You get
- A retrieval assistant over your policies and manuals
- An embeddable widget for your site or intranet
- A citation on every answer and a refusal policy for the rest
- An evaluation set with a pass rate you can re-run
- A cost ceiling, and a note on where AI is not used
Clock
2 to 4 weeks.
Starts with your documents and twenty questions your team actually asks.
Websites & Frontend Design
A site that looks like nobody else’s and loads before they notice.
You get
- Visual direction, type, and motion designed for your business, not a template
- Fast pages that read well on a phone and score well on Lighthouse
- Generated imagery and 3D where it earns its place
- A CMS or content structure your team can edit without me
- Analytics, SEO basics, and a launch checklist
Clock
2 to 4 weeks.
Starts with your current site, your customers, and what you want them to feel.
Custom CRM & Smart Systems
Make the systems you already run talk to each other, and tell you what is happening.
You get
- A CRM or operations layer shaped around your pipeline, not a vendor’s
- Integrations with the tools your team already uses: email, calendars, spreadsheets, accounting, ticketing
- AI where it helps: summaries, triage, next-step suggestions, with a human deciding
- Dashboards for leaders: team workload, throughput, growth, and what is stuck
- Roles, audit trail, exports, and a handoff your next engineer can follow
Clock
3 to 6 weeks.
Starts with a map of your current systems and the three questions you cannot answer today.
Production Readiness Review
You built it fast. I will tell you what breaks first.
You get
- A five-day review of an AI-built app: secrets, database policies, auth, tests, deploy and rollback, cost, maintainability
- A plain-English summary and a ranked list of fixes
- The checklist I used, so you can run it again yourself
- A review, not a penetration test or a certification
- The fee credits toward an Ops Tool Build or an Assistant if I do the fixes
Clock
5 business days.
Starts with read access to the repo and the hosting account.
Keep-It-Running
Monthly fixes, upgrades, and a human who answers.
You get
- Monitoring and a next-business-day response window
- Dependency and model upgrades
- Small changes as your process changes
- A short monthly note on what changed and what to watch
- Pause or cancel any month
Clock
Monthly.
Follows a build. It is not sold on its own.
How I think
THE PART
AFTER “IT WORKS”
Getting a demo to work is the beginning.
Earning trust is the engineering.
01It never guesses.
Rudy answers from a verified corpus or says it does not know. A build gate blocks release when an emergency number cannot be verified.
Inside Rudy’s trust architecture02The rules live in code.
When a license or compliance duty is on the line, the logic is deterministic. The model can explain a decision. It does not get to make the rules.
Explore the Shockwatch sandbox03Every finding becomes a test.
301 adversarial checks on Rudy. 114 tests on Shockwatch. 32 on the inspection tracker. The bugs I find become the checks that stay.
See the inspection tracker04Sometimes the right AI is no AI.
The inspection PDFs already had clean text layers. A deterministic parser replaced the vision model: exact fields, no model rate limits, and no monthly extraction bill.
Read the engineering decisionBuild it. Try to break it. Make the fix permanent.
Experience
Where I’ve
been
From the first prototype to the system people depend on.
Read the full résuméJul 2026 – present
Current roleData Systems & Cloud Infrastructure Engineer
Cal Poly EHS & Risk Management
San Luis Obispo, CA
Build and run the safety and risk tools a campus depends on: a student-facing safety assistant, inspection and fall-protection dashboards, and an enterprise risk platform, all in daily use at Cal Poly and extending to the Solano campus next and Cal Maritime after that.
Jul 2025 – Jun 2026
AI/ML Intern
Cal Poly EHS
San Luis Obispo, CA
Architected a RAG pipeline on AWS Lambda and Bedrock with a Mitigation Strategy Engine that enforces ISO 31000 business logic over LLM output.
Jun 2025 – Jan 2026
Risk Management Intern
Cal Poly Risk Management
San Luis Obispo, CA
Clustered a decade of incident and claims data to isolate loss drivers invisible to manual auditing.
Jun 2024 – Sep 2024
AI Digital Transformation Intern
Meera Services
Remote
Embedded as the forward-deployed engineer between the company and the stakeholders using its billing system, from requirements to post-launch support.
Class of 2026B.S. Computer Engineering, Cal Poly, San Luis Obispo
Four years of Learn by Doing, finished in June 2026 while shipping the tools above. Capstone: Mesosphere, a LoRa mesh for Project OWL, built by a team of four. Senior project: The Sentinel Protocol.
The toolkit.
The right tool for the problem. From retrieval pipelines to the hardware at the edge.
- AI, LLMs & agents
- LLM agentsRAG & retrieval tuningAmazon BedrockClaude APIModel evaluationRed teamingPrompt engineeringNIST AI RMF
- ML & data
- PythonPyTorchScikit-LearnpgvectorAnomaly detectionComputer vision (YOLO)pandasSQL
- Cloud & backend
- AWS LambdaDynamoDBRDSComprehendVercelSupabaseNeon PostgresNode.jsDocker
- Frontend
- TypeScriptReactAstroTailwind CSSThree.js / WebGLGSAPAccessibility (WCAG 2.2)
- Embedded & hardware
- C / C++LoRaWAN meshNMEA & CAN telemetryIMU sensor fusionEdge AI (Jetson)Analog circuit design
Have a hard problem?
Open to AI engineering roles
and a few good projects.