LLM & agentic systems
Multi-agent workflows, MCP servers and prompt systems that run unattended and stay on task.
- LangGraph
- MCP
- Prompt design
Open to remote AI roles
Hi, I’m Aparna Sunil
I build production AI systems with APIs, MCP, LangGraph, RAG and prompt workflows on AWS.
Production AI, from the model to the service around it.
Multi-agent workflows, MCP servers and prompt systems that run unattended and stay on task.
Ingestion pipelines, vector search and grounded answers with hallucination checks built in.
Serverless services that hold up with many customers at once. APIs, Lambda, DynamoDB and SQS.
Connected AI services and automations built with n8n, Python and model APIs.
I’m an AI engineer based in Kochi. My work sits where models meet infrastructure: the contracts between services, the guards in front of generation, and the prompts that have to behave when nobody is watching.
I came to this through physics, then a master’s in data analytics, then a run of end-to-end projects built to be used, not demoed. The part I like most is when “works on my machine” becomes “works for every customer at once”.
How I work
Backend and AI engineering on a multi-tenant content platform: MCP server, compliance engine, knowledge-base pipeline and prompt systems.
Production systems from my work as an AI engineer.
Project
Built backend and AI systems for a multi-tenant content automation platform on AWS. My work covered MCP integrations, compliance workflows, knowledge-base ingestion, scheduling safeguards, and production prompt systems.
What I built
Built end to end, on my own time.
Agentic job assistant: scrapes roles, matches them semantically, finds skill gaps and tailors the resume. Five LangGraph agents.
LangGraph · Streamlit · SQLite · Gemini
GitHubWorkflow-driven RAG chatbot for dermatology questions. n8n pipelines, pgvector search, Flask front end, fully Dockerised.
n8n · Flask · PostgreSQL · pgvector · Groq
GitHubGrounded answers over university PDFs. Reinforcement learning tunes the context size per query; NLI checks catch hallucinations.
Transformers · ChromaDB · Streamlit · Gemini
GitHubGraph convolutional network over credit-card transaction graphs, surfacing the connections that make a transaction suspicious.
PyTorch Geometric · KNN graphs
GitHubDescribe a situation in plain words and get the relevant IPC sections back, summarised for the public.
Sentence Transformers · DistilBART · Streamlit
GitHubWhat I reach for, grouped by the job.
I’m open to remote AI roles where practical systems reach real users.
aparnassunil23@gmail.com