Priyanka Kumari — AI / ML Engineer
I build agentic pipelines, RAG systems and the backend engineering that keeps them correct under real traffic — from a research agent that cites all 13 of its sources to a booking API that has never sold the same ticket twice.
Every project below ships with a build log: the decisions, the trade-offs, and why each one was made.
01 agent = Pipeline(sources=13)
02 report = agent.research(topic)
03 assert report.every_claim_cited
// plan → research → verify → ship
01 — SELECTED WORK
From agentic research to computer vision and high-concurrency APIs. Each project has a build log — the decisions behind the system and why they were made — alongside a walkthrough of how to run it.
GENAI · VIDEO · OPEN-WEIGHTS
A creator dashboard that turns a goal into a finished vertical video — trends, research, script, frames, voice and an original music bed. Every stage runs on open-weights models, and every stage reports whether a model actually ran or a template stood in.
GENAI · AGENTS · RESEARCH
An agentic pipeline that researches any topic across 13 verified sources, cross-checks every claim, scores confidence and writes a fully-cited report — then voices it as a podcast.
BACKEND · CONCURRENCY
A flash-sale booking API built to survive hundreds of buyers hitting the last ticket at once — double-booking ruled out by row-level locking, not by hopeful timing.
COMPUTER VISION · REAL-TIME
Eight hand gestures, finger counting across two hands and seven-class emotion detection from one webcam feed — threaded capture and temporal smoothing keep it flicker-free.
NLP · DEPLOYMENT
Transformer models handle grammar, TextBlob handles spelling, and Docker removes the environment problem — docker run is the whole install procedure.
MACHINE LEARNING · NLP
An end-to-end classifier taken from notebook to browser: cleaned text, TF-IDF features, 98% test accuracy, shipped as a Streamlit app that reads PDFs.
RECOMMENDATION · ML
A content-based recommender over ~5,000 TMDB titles. The similarity matrix is precomputed, so a recommendation is a lookup rather than a calculation.
02 — TECHNICAL TOOLKIT
Connecting models, data, APIs and product experiences into systems that are practical to use and easier to evolve.
Python · SQL · JavaScript
Scikit-learn · Feature Engineering · Classification · Recommendation
LLMs · Agentic AI · RAG · LangChain · LangGraph · Hugging Face
RNN & CNN · OpenCV · MediaPipe · DeepFace · Real-time inference
FastAPI · Flask · PostgreSQL · Redis · Docker · Streamlit
Git · GitHub · Testing · Documentation · Problem Solving
03 — ABOUT
I'm Priyanka Kumari, an AI/ML Engineer focused on building intelligent applications that solve real problems.
My work spans machine learning, NLP, computer vision, Generative AI and backend engineering. I enjoy moving from an ambiguous problem to a working system — understanding the data, choosing the right approach, building the interface and making the result usable.
What I care about most is the part that comes after the model works: making it reproducible, making it fast enough, and making it something another engineer can pick up and extend. Every project in the section above has a build log explaining the decisions behind it, because the reasoning is usually more interesting than the result.
I’m especially interested in AI agents, production-ready ML applications and intelligent automation.
04 — EXPERIENCE
ShipCube · Remote
ShipCube · Remote
Precision Electronics Limited · India
IIMT University
05 — CONTACT
Open to conversations around AI/ML engineering, Generative AI, intelligent automation and challenging product problems.