Hi, I'm Rohit.
A Founding AI Engineer.
I build and scale production AI systems — agentic workflows, voice AI with sub-500ms speech-to-speech, and RAG architectures used by thousands.
Working remotely with teams building AI-native products.
Primarily focused on the AI + full-stack ecosystem, but always eager to explore and adopt what best solves the problem.
Python
TypeScript
React
Next.js
FastAPI
Node.js
PyTorch
TensorFlow
LangChain
OpenAI
GraphQL
Redis
Docker
Kubernetes
AWS
GCP
Tailwind
VercelPythonTypeScriptReactNext.jsFastAPINode.jsPyTorchTensorFlowLangChainOpenAIGraphQLRedisDockerKubernetesAWSGCPTailwindVercelWork & education.
A Founding AI Engineer who's shipped 10+ enterprise-grade agents, taken a platform 0→1,000+ users, and obsesses over the details between demo and production.
- Built a full-stack AI document-generation platform (Next.js, FastAPI) that turns RFPs into citation-grounded drafts, saving 85%+ of writer time.
- Engineered a PDF/DOCX ingestion + OCR pipeline with Docling, OpenAI embeddings, and Pinecone — 280+ docs at 94.3% clean extraction.
- Designed a 3-stage evidence-matching pipeline with LangSmith evals reaching 96.2% claim traceability.
- Shipped a reviewer workflow with SSE streaming, RBAC, audit logging — 97.4% generation success rate.
- Shipped ApplyLoom 0→1 — Next.js + Supabase platform scaling to 1,000+ users, plus a Chrome extension across LinkedIn, Indeed and more.
- Built a real-time mock-interview engine with AI personas and A/B-tested intent routing over LiveKit at sub-300ms latency.
- Fine-tuned Llama 3.3 70B with SFT + LoRA on GCP Model Garden — 90% lower LLM cost, p95 latency 2s.
- Developed a low-latency Voice AI agent on Twilio Media Streams + Amazon Transcribe — sub-800ms turn-taking.
- Orchestrated an outbound outreach engine on SQS + Lambda, scaling candidate engagement 5x.
Milestones in the building journey.
Each project marks a step forward — shipping real systems that solve latency, grounding, and scale problems.
Atimuss Flow
Local-first personal voice agent that lives entirely on device with AES encryption. Sub-500ms speech-to-speech, 159 WPM capture — 4x faster than typing.
ApplyLoom
Job-search platform + Chrome extension scaled to 1,000+ users. Mock-interview engine with AI personas over LiveKit at sub-300ms latency.
CellaNova RFP Platform
Citation-grounded proposal drafting with a 3-stage retrieval + refinement pipeline. 96.2% claim traceability, 97.4% generation success rate.
Let's build something real.
Have an AI product idea or an agentic workflow that needs to reach production? I'd love to hear about it.