Capture the useful corners of the web. Organize and search them locally, protect sensitive collections, and carry your library through explicit backups.
An automated path from scattered job feeds to normalized records, skill signals and recurring reports. The deliverable is a workflow that leaves artifacts.
Vessel tracks pass through three different lenses: formal protocol checks, fuzzy risk and relationship graphs. An explainable maritime anomaly simulation.
Listening beneath the rails. A team-built vibration-monitoring prototype connects piezoelectric sensors, edge analysis and a wireless gateway to a remote dashboard.
See the vehicle. Change the beam. A team-built camera and Raspberry Pi prototype turns an oncoming-traffic signal into a physical high-to-low headlight transition.
Hardware integrationComputer visionResearch
ECS team project · presented at SETCON
SCROLL TO UNFOLD12 / 12
THE EXPLORATION CONTINUES / 13
Other directions. The same curiosity.
More work across products, systems and experiments. Open a project for its scope and tools.
13Market MayaAGENT-BASED SIMULATION+
Trader archetypes, a price-time limit-order book and network rumor propagation, with experiment replay.
An occasion, a mood, a budget. Natural-language intent parsing and TF-IDF retrieval find candidates; a constrained bundling engine brings them together into a coordinated outfit.
Turn sizing uncertainty into an explanation. Body measurements, fit preferences and review signals inform a size recommendation, with confidence scoring and a reverse-logistics simulator.
Find the trade-off between clearing stock and protecting margin. Inventory age, sales velocity and demand elasticity guide constrained markdowns, with financial scenarios to explore their effects.
I build systems that connect. From the model to the interface, the details belong together.
I’m Shiven Patro, a Computer Science undergraduate at VIT-AP. My work moves between full-stack products, AI systems and research: from healthcare software at Cordly AI to federated learning research at NIT Tiruchirappalli. I like understanding a problem deeply enough to build something useful around it.
My favorite part is the in-between: where a rough idea becomes a working prototype, and a working prototype becomes something you want to spend time with.
Explored PyTorch / FastAPI / FedAvg training with failure-aware coordination and ablation testing.
Investigated a decentralized evolution using peer discovery and synchronization. Local data remains at the participating nodes; model updates travel between them.
Two related manuscripts were reported as under submission. Publication and acceptance are not asserted here.
Local trainingModel updatesPeer coordination
COMPLETED INTERNSHIP · COLLABORATIVE RESEARCH
Full-stack development for the healthcare vertical, connecting client requirements to working product features.
Worked across React, Node.js, Supabase and API integration.
Resolved full-stack issues and improved database queries and backend behavior.
Translated client needs into technical specifications and product changes.
Client needProduct + APIDelivery
FORMER INTERNSHIP
An AI internship in my engineering journey, alongside a continuing interest in usable machine-learning systems.
Listed as a former AI internship on my professional profile. Project-specific deliverables are not presented without a public record.
ExploreImplementReflect
FORMER INTERNSHIP
Natural-language processing and review analysis, connecting classification with interpretable trends.
Worked with Python, Pandas, NLTK and Scikit-learn on review classification.
Used visual analysis to communicate patterns and support product discussions.
ReviewsNLP pipelineInsights
FORMER INTERNSHIP
Computer Science and Engineering, with independent work spanning software systems, machine learning and applied research.
Developing a foundation in algorithms, databases, software engineering and computer networks.
Applying those foundations through research, hackathons and complete project builds.
FoundationsApplied workResearch
UNDERGRADUATE · EXPECTED 2027
05 / MOMENTS THAT MATTERCREDENTIALS & TEAM MILESTONES
Small wins. Forward momentum.
The milestones are nice. What they make possible is better.
01 / COMPUTER SCIENCE
Harvard CS50x
Completed CS50x in 2025, with Focus Booster Extension as my final project.
Details & evidence +
A course credential in computer science, covering problem sets and a final software project.
Selected for Amazon’s 2025 ML learning program, spanning deep learning, generative AI and reinforcement learning.
Details & evidence +
Attended the virtual program across four weekends in August 2025, learning with Amazon scientists. The letter of acknowledgement was issued on 10 September 2025.
Top 8 finalist with Team Mechanoids at Coimbatore Institute of Technology’s national hackathon.
Details & evidence +
Represented VIT-AP at the Qubitron 2025 finals on 15 March 2025. The organizer’s email names Team Mechanoids among the top eight finalist teams; the participation certificate records the event and date.
No brief. No right answer. Just curiosity, followed somewhere.
X: POSSIBILITY Y: A LITTLE CURIOSITY
++
PLAY
Drag sideways to pull the letters and release to let them spring back. Arrow keys pull; releasing a key returns the letters. Escape rests the artwork.
Drag sideways to tilt the sculpture and release to let it spring back. Arrow keys tilt; releasing a key returns it. Escape rests the artwork.
A LITTLE PLAYGROUND FOR THE POSSIBLE. LIVE EXPERIMENT
DRAG · RELEASE · ARROW KEYS ↗
SMALL EXPLORATIONS. UNEXPECTED CONNECTIONS.Some of the best ideas start with “what if?”
07 / RESEARCH & DEEPER QUESTIONSQUESTIONS, METHODS & EVIDENCE
01 / OBSERVE?
02 / CONNECT↗
RESEARCH NOTES / 001Learning beyond a single node.
LOCAL DATA × SHARED LEARNING
A QUESTION IS A GOOD PLACE TO START.
EXPLORATION IS A PRACTICE.
THE RESEARCH DESK
Stay with the question.
FEDERATED LEARNING ↗
Learning together, without pooling the data.
A research internship exploring drug–target interaction prediction: from a central coordinator to peer-based learning and failure-aware synchronization.
TWO MANUSCRIPTS REPORTED UNDER SUBMISSION PUBLICATION NOT ASSERTED
ON IEEE XPLORE / TQCEBT 2026↗A SEPARATE RESEARCH DIRECTION
THE QUESTION BEHIND THE PAPER
With less data, what does quantum change?
When medical imaging data is scarce, what does the quantum component contribute to a learning system? A question of inductive bias, explored through a twin-architecture benchmark.
Isolating Quantum Inductive Bias: A Twin-Architecture Benchmark for Data-Scarce Medical Imaging
