Sreenaina Koujala

Full Stack & AI Developer

Currently developing

B.S. Honors Computer Science

Personalized, AI-Generated Learning Platform | Vanilla JS, Handlebars, Vite, Supabase, Deno, Anthropic API

Created study spaces that let students input what they want to learn, how they learn best, and optional material, which the app uses to generate a personalized, interactive workspace of cognitive-science-backed widgets (active-recall concept maps, a Feynman-based evaluator, progressive summarization)
Built an interactive study hub with bulletin boards, a document editor, custom flashcard decks, and more, so a student’s entire study workflow lives in one tool instead of five
Architected the application with vanilla JS + Handlebars + native CSS on Vite over Supabase (Postgres with Row-Level Security, passwordless auth, storage) and Deno edge functions calling the Anthropic API with model tiering (Sonnet for space structuring, Haiku for lightweight widgets), schema-validated JSON output, and pgvector retrieval over embedded user sources

HIPAA-Aligned Perimenopause Platform | React Native, Expo, Supabase, PostgreSQL, RAG | In Development

Architected a full-stack React Native/Expo + Supabase app with HIPAA-equivalent standards under the FTC Health Breach Notification Rule (Postgres Row-Level Security scoped per user on every PHI table, append-only PHI access audit log, synthetic-only dev/staging, de-identified research export isolated from primary tables)
Designed a RAG-grounded explanation layer over a clinician-reviewed, version-tracked medical corpus, with hybrid retrieval (pgvector + Postgres full-text, merged via reciprocal rank fusion), per-chunk citations, and escalation rules enforced in code that defer flagged or low-confidence answers to a licensed clinician
Built features women deserve and need: symptom and cycle tracking with useful patterns for irregular cycles, PCOS, and post-hysterectomy users where standard staging criteria break down, plain-language lab result explanations, and doctor-visit-note generation with predictions computed deterministically (recency-weighted rolling mean + variance, confidence shown in-UI)

Full-Stack Developer | May 2018 – Present | Ashburn, VA

Developed ZazUI, a production data visualization platform deployed across 10+ federal government programs; rebuilt a high-volume data grid with ES6 and the Intersection Observer API, cutting load times 75% for 100K+ row datasets and eliminating scroll-induced UI failures
Architecting an agentic widget-generation system that reasons over a component schema and emits production-ready UI code from natural language specifications, reducing widget development time 80% (~5 days to ~5 minutes per widget)

Full-Stack & AI Engineer (Co-op) | Sep 2025 – May 2026 | Fairfax, VA

Architected BuilTING, an 82,000-line production system on AWS GovCloud, converting unstructured building documentation (PDFs, blueprints, AutoCAD DXF, photos) into IFC4-compliant BIM models openable in Revit, fully automating a manual engineering workflow end-to-end
Designed a seven-stage serverless pipeline (Step Functions, DynamoDB, S3, ECR, Amazon Bedrock) pairing deterministic DXF parsers with a multimodal vision pipeline for drawing classification, title-block reading, and real-world scale calibration
Built a claims-based provenance model where every extracted fact carries its evidence chain and confidence score; a resolution engine groups competing claims by identity, alias, and spatial proximity and resolves conflicts into tagged observations
Hardened with a dual-validator (Zod + Pydantic) inter-language contract layer, golden tests against live renders, and end-to-end audit logging; custom frontend built with a vanilla-JS SPA and a real-time WebGL IFC viewer to interact with renders

AI & Data Exploitation Intern | Jun 2025 – Aug 2025 | McLean, VA

Built WovenAI, a production AI chat platform (FastAPI, SQLite, AWS Strands SDK) with end-to-end LLM streaming and live markdown re-rendering; architected a pluggable MCP tool registry (HTTP and SSE transports, per-request dynamic loading, and graceful fallback) and authored multiple MCP servers including a SELECT-only sandboxed SQLite server
Replaced multi-hour manual analyst workflows with a governed AI interface built on context engineering, strict JSON schema validation, and scoped tool permissions, reducing data retrieval time by 45%

Software Engineering Intern | Jan 2023 | Reston, VA

Built a custom canvas mini-map scroll-bar (drag-to-scroll and click-to-seek view-port navigation) in React and Konva for a real-time RF spectrogram in the spectrum-analysis tool IQEngine
Migrated React code-base from class components to functional and hooks architecture across multiple signal analysis features; shipped 20+ tasks in a 2-week sprint covering spectrogram UI, custom scrolling, and a full UX overhaul