I embed with Fortune 500 teams to take LLM-powered products from whiteboard to production — RAG, agentic systems, fine-tuned models, and the full stack around them.
LK
I've spent 6+ years embedding directly with enterprise teams to design, build, and support LLM-powered applications end to end — from identifying high-value use cases and prototyping, through full-stack development, cloud deployment, and long-term production support.
My work spans financial services (a 251,000-member credit union) and consumer electronics (JBL/Harman, Epson), where I've shipped secure, grounded AI assistants that real customers use every day. I pair hands-on technical execution — RAG, agentic orchestration, LLM fine-tuning, and cloud engineering — with the stakeholder-facing communication that gets things over the line.
Right now I lead technical strategy and delivery across multiple enterprise AI deployments, mentor implementation engineers, and drive C-suite adoption of AI platforms.
Use-case discovery → prototype → full-stack build → deploy → support.
Hybrid retrieval, rank fusion, cross-encoder re-ranking, agentic routing.
QLoRA / PEFT on open models, synthetic-data pipelines, evaluation.
Python, React, FastAPI, AWS & Azure, CI/CD, production observability.
A selection of LLM applications I've architected and delivered for enterprise clients. Several are live in production today.
Architected and shipped a production banking assistant for a 251,000-member credit union — a RAG knowledge pipeline with source-grounded answers, compliance guardrails (PII redaction, prompt-injection defense), and secure multi-layer JWT/JWE auth on stateless AWS Lambda.
Try the live bot · truliantfcu.orgBuilt an end-to-end product-intelligence platform over JBL/Harman's full catalog: search, recommendations, comparison, and troubleshooting. Fine-tuned Mistral-7B with QLoRA, generated synthetic training data with GPT-4o, and served answers from a FAISS index behind an agentic orchestration layer.
Internal contact-center deploymentDesigned a production Agentic RAG support assistant across 20,000+ products in 19 countries. Ingested 400K+ documents with semantic chunking + metadata, and built hybrid retrieval (dense + BM25) with reciprocal rank fusion, cross-lingual embeddings, and cross-encoder re-ranking.
See it live · epson.comConnected conversational AI to a Genesys contact center and Microsoft Dynamics 365 (via Azure APIM) — automating live work-item creation and handing full conversation context to human agents. Implemented Redis session tracking and resolved a critical go-live routing incident.
Visit · unitedfcu.comRAG chatbot over online-course content with NER-based PII masking for FERPA-compliant handling — lecture transcription, chunking, and retrieval-augmented Q&A.
Trained CNNs to classify flood vs. dry regions on satellite imagery using IU's Big Red 200 HPC cluster, supporting a disaster-response research pipeline.
Lead architecture and delivery of production LLM applications across multiple Fortune 500 deployments; mentor engineers, guide contractors, and drive C-suite adoption of the AI platform.
Delivered Epson's multilingual Agentic RAG at scale and United FCU's contact-center AI + CRM integration — full knowledge stack from ingestion through grounded answer synthesis.
Built an end-to-end RAG chatbot with NER PII masking and trained CNNs for satellite flood-region classification; led lab sections and mentored graduate students in ML.
Built Node.js API adapters for ServiceNow ITSM bots (−40% ticket resolution time), WebSocket live-agent transfer, and Redis chat-queue optimization (−50% latency).
Developed a Django chatbot platform (rule-based + ML bots) and a JavaScript web SDK with JWT authentication and session management.
Indiana University Bloomington
Lovely Professional University, India
Amazon Web Services
I'm open to Forward Deployed / AI Engineer roles (hybrid or relocation). If you're building production AI, I'd love to talk.