class KamalUtla:
location = "Bangalore, India"
education = "B.Tech Engineering Physics, IIT Roorkee (8.4/10)"
currently_building = [
"Healthcare RAG systems with agentic workflows",
"LLM evaluation & testing frameworks",
"Multilingual knowledge platforms",
"Production conversational AI agents"
]
tech_stack = {
"languages": ["Python", "SQL", "TypeScript"],
"llm": ["Gemini", "OpenAI", "Anthropic", "HuggingFace", "vLLM", "LangChain"],
"ml": ["PyTorch", "TensorFlow", "scikit-learn"],
"backend": ["FastAPI", "Flask", "PostgreSQL", "MongoDB"],
"infra": ["Docker", "GCP", "GitHub Actions", "Firebase"]
}
@property
def current_focus(self):
return "Making LLM agents production-ready"| Role | Company | Period | What I Did |
|---|---|---|---|
| AI Engineer | BorderPlus | Nov 2025 — Now | Healthcare knowledge ingestion, agentic pipelines, multilingual RAG with citations, compliance evaluation |
| Data Scientist | Convai | Jun 2024 — Oct 2024 | vLLM model serving, hybrid RAG (27% latency reduction), fine-tuned Llama-3-70B/90B-Vision, WebRTC agents |
| ML Intern | Convai | Feb 2024 — Apr 2024 | LLM personality modeling, synthetic data pipelines, conversational agent evaluation |
| Research Intern | McMaster University | May 2023 — Jul 2023 | Multi-objective optimization for EV recycling (MITACS Globalink Fellow) |
| Project | Description | Tech |
|---|---|---|
| Journee | AI travel planner — personalized itineraries with flights, hotels, and conversational editing | React, FastAPI, Gemini |
| ProBack | LLM evaluation framework — prompt playground, A/B variants, LLM-as-a-Judge | Flask, MongoDB, OpenAI, Anthropic |
| FlirtChat | AI conversation skills trainer — real-time scoring with coaching feedback | Streamlit, Gemini |
| Neural Style Transfer | Transfer artistic styles to photos using VGG-19 | TensorFlow, VGG-19 |
| Swach Bharat Bot | Autonomous cleaning robot with IR obstacle avoidance | Arduino, C |
-> Fine-tuned Llama-3-70B and Llama-3.2-90B-Vision for controllability and factuality
-> Reduced RAG pipeline latency by 27% and memory usage by 3-4x with hybrid retrieval
-> Built multi-provider LLM backends (OpenAI, Anthropic, Gemini, LLaMA) with tool calling
-> Designed LLM-as-a-Judge evaluation pipelines for automated compliance scoring
-> 15% token reduction in production agents through prompt optimization
-> MITACS Globalink Research Fellow at McMaster University, Canada
-> Winner — Qiskit Fall Fest 2021, IIT Roorkee Quantum Hackathon
-> IIT Roorkee Encore Award 2023 for all-round excellence
