I'm an AI/ML Engineer who gets a kick out of taking ideas from zero to production. Not the kind who fine-tunes a model in a notebook and calls it a day. I mean actually shipping things: APIs, cloud deployments, real users, real data.
I've spent the last 4+ years obsessing over the full stack of AI: from training PyTorch models and building RAG pipelines to wiring up Kafka event streams and deploying on GCP and AWS. If it involves LLMs, agents, or real-time ML, I've probably broken it three times and shipped it on the fourth.
Right now I'm building toward roles at companies that actually push the frontier: Anthropic, OpenAI, Google DeepMind. Not because of the hype, but because I genuinely care about where this technology goes.
rohith = {
"location" : "Charlotte, NC",
"education" : "MS Information Technology, University of Cincinnati (GPA: 3.89)",
"cert" : "AWS Solutions Architect Associate",
"currently" : "Building production AI systems. Shipping. Repeating.",
"targeting" : ["Anthropic", "OpenAI", "Google DeepMind", "Meta AI"],
"status" : "Open to AI Engineer | GenAI | LLM | MLOps (US)",
}
A Claude-Code-style terminal agent that reads, edits, and runs your code, every write behind a diff you approve. Built on a provider-agnostic framework, so the same agent runs on Claude for top quality or free on Gemini / Cerebras / Groq / Ollama (--offline strips every network tool for air-gapped coding). The multi-provider design unlocks things a single-vendor agent structurally can't:
- Consensus: run a task across several models in parallel; a judge synthesizes one cross-checked answer
- Dojo: rival models each attempt the same change in isolated git worktrees; a judge picks the best diff
- Kaizen: the agent finds a weakness in its own source, fixes it in a worktree, and keeps the diff only if the test suite passes
Python -> 7-package monorepo -> 1,376 offline tests -> MCP + 200 plugins
|
Medical research is buried in millions of papers. I built an agentic RAG system that searches PubMed in real-time, pulls the right papers, and synthesizes clinical answers using Claude Sonnet, all streamed live to the user.
|
I got curious about how Netflix actually works under the hood. So I built it. A proper two-tower PyTorch model with BPR loss, 4-source candidate generation, LightGBM reranking, and a Kafka event pipeline on Kubernetes.
|
|
I got tired of seeing "Humbled and excited to announce" for the 50th time. So I built a Claude-powered detector that scores LinkedIn posts for corporate cringe (0-100), generates a one-line roast, and rewrites it like a normal human. Because someone had to.
|
A full ML forecasting platform with a proper pipeline: PyTorch NN + LSTM ensemble, 128-dimensional feature engineering, SHAP attributions for explainability, and KL-divergence drift detection that auto-retrains when the distribution shifts. 14 screens on GCP.
|
|
Enterprise fraud detection with LangGraph agents doing the heavy lifting: real-time risk scoring, multi-step reasoning on each transaction, full audit trail on every decision. Deployed on AWS EC2 with Nginx and SSL. Actually production-grade.
|
🚌 MR BusesA full interstate bus booking platform with an AI chatbot that actually knows the routes, schedules, and can help you book, not just answer FAQs. Google OAuth, real-time seat booking, admin dashboard. Live on GCP.
|
"All live" isn't a slogan. A GitHub Action pings every deployment and rewrites this table. Receipts, not vibes.
| System | Status | Response |
|---|---|---|
| RO MedRAG | 🟢 LIVE | 248 ms |
| BullshiftDetector | 🟢 LIVE | 37 ms |
| MR Buses | 🟢 LIVE | 61 ms |
| RO Fraud Detection | 🟢 LIVE | 251 ms |
🤖 Checked automatically every 6 hours by GitHub Actions · last run 2026-06-18 10:46 UTC
- Ronin · pushed 2026-06-18 · Masterless, terminal-native coding agent (Claude Code-style: reads, edits, runs code) f…
- .github · pushed 2026-06-11 · Community health defaults
- agentfaceoff · pushed 2026-06-11 · Live LLM battle arena · same prompt to two models, token-by-token split-screen streamin…
- rovex-ai · pushed 2026-06-11 · Enterprise fraud detection · 4-model ensemble scoring with a LangGraph investigation ag…
- prior-auth-agent · pushed 2026-06-11 · Prior-authorization agent · citation-grounded payer decisions.
flowchart LR
subgraph Data["📥 Data & Retrieval"]
Docs[Docs & APIs] --> VS[(FAISS · pgvector · hybrid search)]
Events[Events] --> Kafka[Kafka] --> CH[(ClickHouse / BigQuery)]
end
subgraph Brain["🧠 Agents & Models"]
VS --> LG[LangGraph: supervisor · planner-executor · HITL]
LG <--> Claude[Claude API]
PT[PyTorch Two-Tower / LSTM] --> LGB[LightGBM Rerank]
end
subgraph Evals["🧪 Evals & Observability"]
LG --> EV[RAGAS · golden datasets · LLM-as-judge]
EV --> LF[Langfuse: traces · cost · drift]
end
subgraph Serve["🚀 Serving & Infra"]
LG --> API[FastAPI + SSE]
LGB --> API
API --> UI[Next.js / React]
API --> Cloud[GCP · AWS · Azure] --> K8s[Kubernetes + Terraform]
end
Languages
LLMs & GenAI
Agents · RAG · Vectors
Evaluation & Observability
ML & DL
Voice & Real-Time
Backend · Cloud · Infra
Data & Frontend
| Credential | Details |
|---|---|
| Claude with Google Cloud's Vertex AI | Anthropic Education · May 2026 |
| Claude 101 | Anthropic · Apr 2026 |
| AI Fluency: Framework & Foundations | Anthropic · Apr 2026 |
| AWS Certified Solutions Architect – Associate | Amazon Web Services · May 2025 (exp. May 2028) |
| MS, Information Technology | University of Cincinnati · GPA 3.89 · Dec 2024 |
| Deep Learning Specialization | Andrew Ng / DeepLearning.AI · 2024 |
| LangChain & LLM Agents for Production | DeepLearning.AI · 2024 |
🤖 The portfolio has an AI narrator and an "Ask Rohith" chat, and it rewrites itself depending on whether you're a recruiter, an engineer, or just curious. Go say hi.