LLMs
GPT-4, Claude, LLaMA
Hi, I'm
Dharani Chalichama
I build LLM, RAG, and multi-agent systems that hold up in production, plus the full-stack products around them.
Get To Know More
4+ years
Generative AI & Full-Stack Engineering
M.S. Computer Science
University of North Texas, 2025
B.E. Computer Science
KL University, 2023
I'm a Generative AI Engineer at BN Associates, where I've spent the last two and a half years as lead engineer for an enterprise healthcare client. I first built their full-stack product (React/Next.js web, React Native mobile, Python services), and today I own the LLM, RAG, and agent work that runs on top of it.
I enjoy taking a business problem all the way to a service that's measured and monitored in production, using LangGraph agents, custom MCP servers, and grounded RAG. In healthcare every agent action has to be traceable, so I care about evaluation and governance as much as the model itself.
Explore My
BN Associates Inc.
BN Associates Inc.
Infosys
GPT-4, Claude, LLaMA
LangChain, CrewAI
Custom servers
Pinecone, FAISS
LoRA, BERT
Guardrails, HITL
FastAPI, Django
Next.js, Native
JavaScript
REST APIs
GraphQL, gRPC
PostgreSQL, Redis
Bedrock, SageMaker
OpenAI, Vertex AI
PyTorch, TensorFlow
Docker, Kubernetes
Terraform, Jenkins
OpenTelemetry, ELK
Browse My Recent
Stateful LangGraph agents that plan, check each other's work, and carry multi-step service requests to completion. Low-confidence cases go to a human before any final action.
Retrieval over approved enterprise knowledge, with a Neo4j knowledge graph for questions that span users, documents, roles, and organizations.
About 30% fewer hallucination-driven errors
Custom MCP servers give agents scoped access to ServiceNow, Salesforce, internal databases, and email. Agents read incoming messages, pull out what matters, and draft replies for review.
A voice-enabled healthcare assistant that reuses the chat assistant's LLM and RAG layer, so there is only one stack to maintain.
OpenCV cleanup, OCR, and a CNN classifier behind a secure upload flow. The same flow feeds the embedding pipeline.
About 40% less manual review time
BERT models fine-tuned for intent classification and entity recognition on healthcare requests, then exported to ONNX and quantized for fast serving.
About 25% higher NLU accuracy
Client work is described in general terms to respect confidentiality.
Full-stack web app built with React, Node.js, and MongoDB.
This site: responsive HTML, CSS, and JavaScript, hosted on Netlify.
Experiments, coursework, and side projects.
Credentials & Learning
Google · Coursera · Nov 2026
Google · Coursera · Nov 2026
Google · Coursera · Nov 2026
IBM · Coursera · Nov 2025
IBM · Coursera · Nov 2025
IBM · Coursera · Oct 2025
Lowe's Companies, Inc. · via Unstop
Get in Touch
Based in Dallas, TX and open to relocation, hybrid, or remote roles. The quickest way to reach me is email or LinkedIn.