Eastern University
MBA and MS Data Science candidate, connecting product decisions with statistics, evaluation, data analysis, and business context.
AI product designer & developer
Prasiddha
Karki
01 / Profile
Evidence-first systems that retrieve, apply transparent logic, and keep the decision human.
I am a Texas-based AI product designer and developer building end-to-end workflows around public data, retrieval, transparent rules, evaluation, and human control. My independent portfolio focuses on making sources, system state, uncertainty, limitations, and the decision that still belongs to a person visible in the interface.
I am an MBA and MS Data Science candidate at Eastern University, with experience in insurance and healthcare administration informing how I frame evidence, operational constraints, and human judgment.
I am available for focused product design, rapid prototyping, and AI system collaboration with teams working globally.
02 / Background
Product, data, and domain context
Insurance + healthcare administration
Experience in evidence-sensitive, operational settings where workflow clarity, privacy, and human review matter.
03 / Capabilities
End-to-end product thinking
Retrieval & knowledge
Grounded interfaces over public, document, and structured sources with provenance, freshness, permissions, and missing-data states designed into the product.
Agentic workflow design · practice direction
Planning tool-using flows with visible state, scoped actions, retries, approval gates, and fallbacks. The public portfolio demonstrates the supporting retrieval and interaction foundations, not a deployed autonomous agent.
Multimodal product UX
Interfaces that combine text, documents, images, maps, and structured inputs while keeping the underlying system legible.
Evaluation & safety
Task-based evaluation surfaces, failure analysis, deterministic baselines, safeguards, review paths, and honest claims about uncertainty.
04 / Independent product lab
Six working products
Independent explorations, not client engagements or production-performance claims.
Public-data intelligence
Voltline
Retrieves EIA-930 hourly submissions for Texas, validates and aligns demand with forecast and adjacent signals, then explains visible variance through deterministic rules without presenting an operational grid alert.
Environmental decision support
HeatSignal
Resolves worldwide locations, retrieves latest-available weather and modeled air-quality data, calculates the NWS heat index when applicable, and keeps observation times and official guidance visible.
Federal-record retrieval
FieldLens
Validates and decodes a VIN with NHTSA data, retrieves matching recall campaigns, surfaces urgent warnings, and separates federal safety records from title and history checks handled by approved providers.
Website evidence preflight
SceneCraft
Safely retrieves bounded public evidence, combines page signals with RDAP, malware-filter DNS, and available header context, and returns clearly separated full or partial results instead of scoring unknown evidence.
Geospatial retrieval & ranking
PlateScout
Uses Nominatim, Overpass, and OpenStreetMap tags to find nearby food places and rank declared cuisine fit, distance, and source completeness without inventing ratings, prices, or popularity.
Browser-local document tool
PaperPatch
Renders and edits visitor-selected PDFs inside the browser with PDF.js and pdf-lib, exports visible additions locally, and distinguishes visual signatures and cover-ups from identity proof or secure redaction.
05 / Approach
Design the loop, not just the prompt
- 01
Frame
Define the user decision, source of truth, acceptable failure, and what must remain human.
- 02
Ground
Connect trusted context, tools, permissions, citations, and provenance.
- 03
Prototype
Build the smallest end-to-end workflow and expose its state in the interface.
- 04
Evaluate
Test quality, latency, safety, and failure modes against representative tasks and a non-AI baseline.
- 05
Improve
Add feedback loops, escalation, monitoring, and versioned changes before expanding autonomy.
06 / Availability
Focused collaboration around a real product problem.
Current portfolio services include AI product diagnostics, working prototype sprints, and system-and-evaluation partnerships. The starting point is the user decision, the evidence required, and what cannot be allowed to fail.