Ananya   Shrimali

Project Overview

Type: Master's Thesis — Computational Building Performance
Duration: Spring 2026 (Ongoing)
Location: Pittsburgh, PA — Gulf Tower Case Study
Advisor: Prof. Dr. Yiqun Pan
Scope: Pre-Design & Schematic Design Phase

Context: Office Floor Plate Typologies

Office building floor plates have evolved through several shapes over time. While I, U, L, and H typologies have seen successful residential conversions, deep rectangular floor plates remain the most challenging — and are the focus of this thesis.

Case Study: Gulf Tower, Pittsburgh

Art Deco high-rise (1932) with a rectangular deep floor plate (~192 ft × 132 ft) in Downtown Pittsburgh, with floor plate depths of ~60 ft on one side and ~45 ft on the other. Currently proposed for office-to-residential conversion with funded development backing.

Problem Statement

One of the primary technical barriers to converting deep floor plate office buildings into residential spaces is daylighting performance in interior dark zones. Office buildings feature deep floor plates and central service cores that create areas unsuitable for residential living. Achieving residential standards (LEED v4 sDA ≥55%) is difficult in zones more than 30–35 feet from exterior windows.

Ref: Jiyoon Lee in How to Tailor 50-story Office, Harvard GSD 2023

Methodology — Phase 1 (Completed)

Developed an early-stage parametric workflow to understand the maximum livable daylight threshold achievable through combinations of unit depth and facade transparency. Unlike typical office daylighting studies that simulate the entire floor plate as one zone, this approach subdivides into individual dwelling units — each with different depth, orientation, ceiling height, and WWR.

Methodology — Phase 2 (Ongoing)

Developed the O2R Daylight Advisor — an interactive decision framework that ingests HB-Radiance simulation output and lets architects explore multi-objective daylight optimization in real time. The tool has four integrated views:

• Feasibility Assessment — Set inherited building constraints (orientation, min WWR, max unit depth) and instantly see which configurations achieve LEED v4 compliance (sDA ≥55%, ASE ≤10%)
• Cost-Tier Comparison — Interventions grouped by renovation cost to find the point of diminishing returns where higher glazing spend no longer improves sDA
• Parallel Coordinates — Brush-filter across all 10 input parameters simultaneously to discover non-obvious facade transparency combinations
• Orientation Advisor — Facades ranked by average sDA with best-performing configuration for each, guiding which side to prioritize for habitable rooms

Methodology — Phase 3 (Final Steps)

Based on tool results, creating a schematic residential unit layout along with facade transparency recommendations for optimum daylight performance.

Tools & Technologies

Rhino, Grasshopper, RhinoPython, HB-Radiance, Python, Interactive Data Visualization

Overview

Full-stack web application serving 200+ weekly participants with real-time attendance tracking, point management, and prize redemption.

Tools

React, JavaScript, Google Apps Script, Google Sheets API, XLSX.js

Overview

Winner of Harvard HPAIR Impact Challenge 2025. ML model for fire-spread prediction using geospatial data and simulation-based analysis.

Tools

Python, Machine Learning (Scikit-learn), GIS Analysis, Data Visualization

Overview

Interactive data science web application analyzing 100,000+ drug-related data points with geospatial visualizations and time-series analysis.

Tools

Python, Pandas, NumPy, Altair, Pydeck, Streamlit

Project Overview

Building Type: Three-story public library
Location: São Paulo, Brazil
Climate Zone: 2A (Warm, Humid)
Service Life: 50 years
Team: Anisha, Ananya, Kansi, Shruti

Climate Characteristics

High temperature and humidity year-round, especially in summers. DBT comfort range: Winter (71°F–78°F), Summer (76°F–82°F). High humidity levels throughout the year drive the need for dedicated dehumidification strategies.

Primary HVAC System: VRF with DOAS

The Variable Refrigerant Flow (VRF) system with Dedicated Outdoor Air System (DOAS) was selected for its flexibility for future space modifications, individual zone control for occupant comfort, and quiet operation essential for a library environment — achieving 41–66% energy cost reduction compared to conventional systems.

Zoning Strategy

Space-based zoning across reading areas, study zones, circulation, and administrative spaces. VRF outdoor unit placement optimized for system efficiency with multiple indoor units distributed per functional zone.

Additional Green Features

Green roof potential, bio-wall integration, and solar-assisted cooling enhancement complement the core HVAC strategy for a holistic low-carbon building approach.

Tools & Technologies

ClimateStudio, Energy Modeling, ASHRAE Standards, PVWatts, VRF-DOAS Design

Overview

Comprehensive HVAC system selection and energy use estimation for a small coworking office in San Francisco, CA. The building includes a coworking area, private rooms, and shared recreational spaces divided into five thermal zones. The project evaluates four HVAC system types, selects the optimal system through comparative analysis, performs detailed load calculations, sizes an Energy Recovery Ventilator, and quantifies energy savings through a baseline-vs-VRF comparison in Climate Studio.

Team: Ananya Shrimali, Migel Shehu, Visha Ahmad
Course: 48-798 Environmental Performance Simulation
Location: San Francisco, CA (mild climate, 250+ sunny days/year)
Software: Climate Studio, RenewAire CORES Calculator

Part A — HVAC System Determination

Four systems were evaluated for the small office: Ductless Mini-Split, VAV, Geothermal Heat Pump, and Water Source Heat Pump (WSHP). Each was assessed on energy efficiency, cost, maintenance, occupant comfort, and scalability.

Part B — Thermal Analysis & Load Calculations

Thermal analysis conducted in Climate Studio using San Francisco International Airport weather data. A baseline ASHRAE model was modified for better energy efficiency. Peak heating and cooling loads were calculated for all five thermal zones.

Part C — Energy Recovery Ventilator (ERV)

ERV sized using the RenewAire CORES calculator with ASHRAE 90.1 Standards and 62-2016 ventilation standards for the San Francisco location.

Part D — Baseline vs. VRF Energy Comparison

Tools & Technologies

Climate Studio, ASHRAE 90.1 Standards, ASHRAE 62-2016, RenewAire CORES Calculator, VRF System Modeling, Load Calculations, Mitsubishi Electric PUMY-P60

Overview

This project explores how daylight can be orchestrated to evoke emotional memories and recreate nostalgic experiences. The design reimagines a family farmhouse gathering space in Central India, translating personal memories and a carefully curated music playlist into spatial lighting experiences using evidence-based daylighting metrics.

Design Concept

The space recreates cherished childhood memories of summer visits to a family farmhouse, where two families would gather to enjoy ice cream sundaes while listening to a shared playlist. The design hypothesis: daylight quality and distribution can be choreographed to parallel the emotional arc of music, creating spatial "verses," "choruses," and "intros" that correspond to different lighting qualities throughout the room.

Musical Lighting Zones

The design translates a Bollywood-inspired playlist into architectural form, mapping musical characteristics to lighting qualities across three zones:

• Intro Zone (near TV): Dim, dreamy environment below 300 lux — acoustic, gentle intros mapped to serene, dim entry lighting
• Verse Zone (central): Moderately lit transitional space at 300-500 lux — reflective, melodious verses mapped to soft spectral light play
• Chorus Zone (near balcony): Bright, warm space at 1000-3000 lux — heartfelt, warm choruses mapped to comforting light with views to nature

Design Narrative

Material palette includes exposed concrete, warm wood screens, Persian rugs, and strategic glazing oriented toward greenery, reinforcing the indoor-outdoor connection central to the farmhouse memory. The compact single-room approach reflects authentic memory — the families always gathered together in one space, honoring the intimacy and togetherness central to the nostalgic experience.

Tools & Technologies

ClimateStudio, Rhino, Grasshopper, Radiance, VR Visualization

Overview

A spatial analysis of industrial burden, race, income, and poverty across Louisiana parishes. The 85-mile corridor along the Mississippi River between Baton Rouge and New Orleans contains over 150 petrochemical facilities concentrated in predominantly Black, low-income communities. This project investigates how historical land use decisions, industrial zoning practices, and racialized housing patterns created disproportionate environmental exposure.

Demographic Analysis

Choropleth mapping of racial composition reveals that parishes with higher Black population percentages overlap with regions of intense industrial activity. Within Cancer Alley, East Baton Rouge, St. John the Baptist, St. James, and Iberville show noticeably elevated percentages, underscoring environmental justice concerns.

Economic Vulnerability

Lower-income concentrations appear in northeastern and central parishes ($28,000-$40,000), while Cancer Alley parishes show moderate to moderately high income levels but remain surrounded by lower-income regions. Despite major industrial corridors, many Mississippi River communities fall below statewide upper income ranges.

Industrial Burden Analysis

The largest emission sources are heavily clustered along the Mississippi River corridor. Cancer Alley parishes exhibit both high facility density and the largest total-release values statewide, while rural and northern parishes show substantially lower cumulative toxic burden.

Kernel Density Estimation

Hotspot analysis reveals the most intense industrial clusters concentrated in the Mississippi River corridor, particularly in Ascension, St. James, St. John the Baptist, St. Charles, and Jefferson parishes. The spatial alignment of dense facility clusters with Cancer Alley underscores the disproportionate environmental burden carried by this region.

Multivariate Environmental & Social Risk Clustering

Integrated analysis combining industrial burden (mean toxic releases), poverty rate, median household income, and percent Black population across all Louisiana counties produces a five-cluster risk classification. Cancer Alley parishes consistently fall into the highest and moderate risk clusters, confirming that the region is not only an industrial landscape but also where environmental burden and racial inequity intersect most strongly.

Data Sources

Louisiana Counties Boundary Dataset, ACS Race Demographics, ACS Median Household Income, ACS Poverty Variables, US Major Rivers Polyline Dataset, EPA Toxic Release Inventory (TRI)

Tools & Technologies

ArcGIS Pro, ArcGIS StoryMaps, ModelBuilder, Python (ArcPy), Spatial Statistics, Kernel Density Estimation

Overview

"Don't just track actions. See them add up with thousands of others, in real-time." Carb-Off is a conversational AI carbon tracker inspired by Indore's proven behavior change model — the city that went from average cleanliness to #1 in India for 7 consecutive years through visible progress and collective identity.

The Indore Story

In 2016, my hometown Indore was an average city in terms of cleanliness. By 2023, it was the cleanest for the 7th year running. The secret wasn't fancy technology — it was visible progress (everyone could see rankings update in real-time) and collective identity (your actions joined something bigger). 3.5 million people changed behavior using this exact model, sustained for 7+ years.

How It Works

Onboarding: Tell the AI where you live, how you commute, what you eat, and your energy habits. It calculates your personal carbon baseline in 2 minutes.

Logging: Type naturally — "I took the bus today and had a veggie burrito for lunch" — and the AI calculates your savings with tangible equivalencies.

Collective Impact: Watch the live counter tick up as your savings join hundreds of others in real-time.

Why the LLM Is Not a Gimmick

The LLM handles three hard problems: intent parsing ("biked to work" → round trip or one way?), contextual calculation (auto-fills distance from user profile), and ambiguity resolution (asks natural clarifying questions). Critical rule: the LLM classifies and communicates — it NEVER generates carbon numbers. Every number traces to a cited emission factor from EPA 2024, DEFRA 2024, or Poore & Nemecek 2018.

Tech Stack

Backend: FastAPI (Python), Claude Sonnet 4 (Anthropic AI), in-memory database (SQLite-ready)
Frontend: React (vanilla — no build step), mobile-first responsive design, real-time polling
Data: 25+ emission factors across food, transport, home energy, and lifestyle categories

Tools & Technologies

Python, FastAPI, React, Anthropic Claude Sonnet 4, EPA/DEFRA Emission Data

Climate-Driven Design — Miami Small Office

Design iterations based on Climate Studio tools for a small office in Miami, FL. Climate analysis informing form, orientation, and envelope strategies across 3 parametric iterations targeting passive cooling, solar shading, and natural ventilation.

Secondary School Building — Pittsburgh (IESVE Analysis)

Climate Zone: 5A (Pittsburgh)
Software: IESVE (Integrated Environmental Solutions Virtual Environment)
Standards: ASHRAE 90.1-2016, PNNL guidelines
Categories: Envelope geometry, envelope thermal properties, internal loads, operational schedules

Climate Analysis Integration

Tools & Technologies

Climate Studio, IESVE, ASHRAE 90.1-2016, PNNL Guidelines, Parametric OFAT Analysis, Daylighting Metrics (sDA/ASE/DGP)

Overview

Whole-building energy performance simulation using IES-VE 2024 for a secondary school in Pittsburgh, PA (ASHRAE Climate Zone 5A). The study explores how site conditions, weather data, building materials, internal loads, and operational schedules shape energy performance during the design development phase. Starting from a PNNL baseline model, parametric simulations across four categories achieved a 45.45% reduction in site EUI, from 33 to 15 kBtu/ft²/yr.

Software: IESVE (Integrated Environmental Solutions Virtual Environment)
Standards: ASHRAE 90.1-2016, PNNL Secondary School model, LEEDv4 compliance
Team: Team of 4
Instructor: Professor Yiqun Pan

1. Baseline Building Energy Model

The building is a two-floor secondary school prototype from the PNNL model dataset. The first floor includes classrooms arranged in pods, a gym, auditorium, auxiliary gym, kitchen, and cafeteria. The second floor has additional classrooms, offices, and a library media center. Pod-based layout with classrooms in corners and central corridors.

2. Design Alternatives — Parametric Optimization

Four categories tested using One-Factor-At-A-Time (OFAT) methodology: envelope geometry, envelope thermal properties, internal loads, and operational schedules. Each iteration built on the previous best result.

3. Best Design Model Results

4. Comparative Analysis

Tools & Technologies

IESVE 2024, ASHRAE 90.1-2016, PNNL Secondary School Model, LEEDv4 Compliance, Parametric OFAT Analysis, Thermal Zoning, Energy Performance Metrics (EUI)

Project Overview

Type: Thesis Project — Cultural/Memorial Architecture
Duration: 4 Months
Location: Taki, West Bengal, India (India-Bangladesh border on River Ichhamati)
Recognition: Awarded "Best Thesis 2023"
Inspirations: Louis I. Kahn, Charles Correa, Peter Zumthor, Juhani Pallasmaa

Design Philosophy

Architectural storytelling of the 1971 Bangladesh Liberation War through phenomenological design, creating an experiential journey that transitions users through emotions chronologically as historical events unfolded. The museum uses Gordon Cullen's Serial Vision — sequential discovery, moments of surprise, and sensory manipulation through water, light, and shadow.

North Museum Block — Bangladesh Perspective (6 Blocks)

Narrative: Before the War → West Pakistan Dominance → Genocide → Awami League Win → Liberation Attempt → Attempt to Escape
Character: Closed envelope, limited calculated light, dark/calm/cool atmosphere
Materials: Rat-trap bond double-wall system using local brick from neighboring factories

South Museum Block — Indian Perspective (6 Blocks)

Narrative: Indian Support → Jessore Liberation → Upper Hand → Operation Python → Battles → Victory
Character: Open envelope, multiple light sources, bright/warm/energetic atmosphere
Strategy: Progressive liberation through increasing openness

Building Performance & Climate Response

Solar orientation: East-west axis aligned with sunrise/sunset for symbolic "birth of new nation"
Sunpath optimization: Waterbody positioned to create reflections during key times of day
Natural ventilation: Open central space above grade with below-grade exodus tunnel to separate thermal zones
Daylighting: North Block (dark, controlled) vs. South Block (bright, open envelope)
Passive shading: Local bamboo jaali (lattice) creates dynamic shadow patterns — direct sunlight, still water reflection, dynamic water reflections

Light & Shadow Studies

North Block (Dark Narrative): Minimal southern openings, strategic northern placement, orderly shadows, broken mirror reflections showing "shattered lives", jaali creates moving shadows through water reflection.

South Block (Bright Narrative): Maximum southern exposure, large waterbody amplifies light, multiple pause points to combat museum fatigue, energetic warm atmosphere.

Experimental Work: Physical models testing bamboo jaali shadow patterns — direct sunlight effects, still water reflections, dynamic water reflections integrated into museum design.

Serial Vision — Design Methodology

Gordon Cullen's Serial Vision applied: existing view → emerging view → sequential discovery → moments of surprise → sensory manipulation (water sounds without direct visibility). Building heights correspond to emotional intensity of events. Sky captured in architectural frames. Landscape enhanced through reflection.

Tools & Technologies

3D Modeling, Solar Analysis, Perspectival Framing, Rendering, Physical Model Making, Phenomenological Design

Project Overview

Type: Studio Project — Mixed-Use (Commercial + Residential)
Duration: 3 Months
Location: Downtown Kolkata, India
Climate: Hot-Humid Tropical
Recognition: Top 5 in External Jury Critique

Design Concept — The Fourth Dimension

The "Tesseract" metaphor extends architecture beyond 3D (closed and open spaces) into a 4th dimension: intangible culture development through built form. The building acts as a catalyst for community formation — physical networking as architecture's fourth dimension.

Climate-Responsive Form Development

Wind Flow Strategy

Void openings aligned with prevailing wind direction (especially summer months). Natural cooling through accelerated airflow. Inner open spaces shaded most of day throughout year. Reduced mechanical cooling loads through passive ventilation.

Building Program

Community-Building Through Design

Design Psychology

Smaller volumes = better engagement — research shows people interact more in compact spaces. Shared activities = common ground — similar interests create conversation and bonds. Reward-based destinations — coffee shops, gardens as "rewards" for social trips. Multiple engagement points attract diverse residents.

Urban Design — "The Street Isn't Curb to Curb"

Ground level: Setback activation with public architecture on lower levels, breaking the rigid compound wall concept.
"At 4th floor we lose identity with streets" — public spaces on floors 0–3 for street connection, private residences floor 4+ for disconnect from noise.
Marketing through visibility: When people see others enjoying services, they're attracted — better than banners.

Circulation & Privacy

Non-monotonous corridor layouts with private entrance zones per apartment. Visual variety breaks repetition. 4 residential cores + 2 commercial cores (completely separate). Basement provides direct access to residences only — privacy maintained between residential and commercial users. Unique unit placements enable varied pricing flexibility.

Computational Design Process

Iterative optimization: Functional stacking (baseline) → wind-responsive dislocation → climate software validation → porosity refinement → activity space integration → final form.
Tools: Climate analysis software (wind flow + comfort validation), 3D modeling (Rhino/SketchUp), daylighting analysis (shadow studies for void placement), conceptual CFD for Venturi effect validation.

Tools & Technologies

Climate Analysis, CFD (Conceptual), 3D Modeling, Daylighting Analysis, Parametric Form Development, Urban Design

Project Overview

Type: International Competition
Duration: 54 hours (timed competition)
Team: 3 members
Location: Tianducheng, China (Parisian replica / "duplitecture")
Recognition: International Winner — Docexdoce Competition
Publication: Web publication

Problem Statement

Tianducheng is a Chinese suburb replicating Paris (including an Eiffel Tower replica), resulting in ghost town conditions — used only as a photo backdrop with informal, unstructured living, no cultural identity, and a small growing population with zero community engagement. The challenge: create cultural identity through intangible daily activities rather than iconic architecture replication.

Design Philosophy

"Culture does not make people, people make culture" — Chimamanda Ngozi Adichie

Replicating heritage buildings does not equal cultural authenticity. Cultural identity comes from the intangible aspects of daily resident activities. Architecture serves as a catalyst, not the solution. The design responds through a simple, effective grid system with adaptive squares enabling sense of ownership, developing community through three pillars: interaction, collaboration, and creation.

Urban Strategy

Circulation: Grid pathways ensure connectivity with multiple access points to each square. Pedestrian-priority environment with flexible routing.

Landscape: Green infrastructure in every grid with varied topography, water features, and native planting.

Infrastructure: Underground utilities, distributed water/power to each grid. Modular infrastructure allows phased development with sustainable drainage systems.

Tools & Technologies

Urban Design, Adaptive Grid Systems, Community Participation Design, Landscape Architecture, Phased Development Strategy

Project Overview

Type: International Competition
Team: 3 members
Location: Plage Long Beach, Cannes, France
Program: Festival pavilion + pop-up cinema
Recognition: International Top 30 — Archasm Competition
Publication: Web publication

Design Philosophy

"Architecture and filmmaking, both should be able to create a narrative other than mainstream, in order to be incredibly successful."

Architecture and filmmaking are treated as parallel narrative mediums. Three elements of storytelling — Character, Conflict, and Desire — are translated directly into architectural form, creating a pavilion that visitors experience as a story unfolding.

Volumetric Strategy

Volume: Create desire through elevated seating platform.
Spine: Exclude all faces, retain only structural skeleton — maximum visual impact with minimum material.
Skin: Cover select volumes (museum, ramp, café) with tensile fabric for weather protection while maintaining porosity.

Experiential Journey

Approach: Skeletal character visible from distance on beach → Entry: Ground-level showcase space → Ascent: Ramp creates gradual elevation + anticipation → Convergence: Social gathering node at mid-level → Souvenir: Commemorative stop before cinema → Arrival: Elevated cinema seating with ocean view and sunset framing.

Climate Response — Coastal Environment

Sun: Tensile fabric skins provide targeted shade.
Wind: Porous frame allows sea breeze through — natural ventilation.
Rain: Protected areas under fabric skins.
Heat: Elevated platforms catch sea breeze, thermal mass minimized for temporary use.
Views: Beach-facing transparency, city-facing enclosure. Elevated platforms frame ocean + sunset.

Cultural Context

The Cannes Film Festival celebrates the collective effort of creative people. Audio-visual experience requires synchronization of senses. The pavilion becomes a "buildification" of story — spaces transition through drama and sophistication phases, with architecture evoking emotion in parallel to film.

Tools & Technologies

3D Modeling, Structural Design, Tensile Fabric Systems, Temporary Construction, Narrative Architecture, Experiential Design

Overview

Pandemic response architecture addressing the COVID-19 housing crisis for urban poor and migrant workers. During lockdown, millions of migrant workers were stranded in cities — unable to return home, without transportation, spending months in goods carrier trucks and streets with complete loss of income and shelter.

Recognition: Shortlisted — IGBC (Indian Green Building Council)

Why Hexagonal Geometry?

Psychological Benefits: Reduces feeling of "packed closeness," provides more atmosphere for 24/7 indoor stay, psychologically more comfortable shape, less claustrophobic than rectangular boxes.

Functional Benefits: Optimized daylight exposure with more facade area per unit, better ventilation from multiple facade orientations, spatial efficiency fitting more units in same area, circulation flexibility with no dead ends, and layout that promotes socialization.

Environmental Benefits: Maximum outer facade area for daylighting, better natural ventilation from multiple angles, reduced mechanical cooling needs, and integration of community gardens between units.

Module Design — The Kit

Flat-Pack Delivery System:

• Structural frame: Timber/metal interlocking members
• Floor plates: Modular flooring sections
• Wall panels: Shutters/covering with insulation
• Roof components: Weatherproof covering
• Joineries: Connectors, fasteners, hardware

Materials: Paper, wood, straw (sustainable, affordable). Fabric walls for flexibility, natural insulation (straw, cardboard), residents can customize with local materials.

Assembly Process

Step 1 — Interlocking: Frame members slot together using joinery system requiring no specialized tools. Residents can self-assemble with basic instruction.

Step 2 — Finishing: Structure + covering material put over frame. Residents customize screens/walls. Personalization creates sense of home.

Step 3 — Combining: Units combine in various honeycomb geometries with infinite arrangement possibilities. Shared walls reduce material needs, community develops organically.

Urban Integration

High-Density Cities (Mumbai): Vertical towers in gaps between buildings, near parking, in unused lots, on underutilized rooftops of existing buildings.

Semi-Urban Areas: Horizontal clusters with community spaces.

Rural Areas: Scattered individual units near work sites.

Tower Configuration: Ground level (community kitchen, storage), mid-levels (residential units), roof (garden, gathering space, water collection).

Community Spaces Integration

Between units: community gardens, children's play areas, shared cooking facilities, learning/skill-sharing spaces, storage.

"Tower as Second Place": Learning opportunities, skill development workshops, social support network, mental health through community connection.

Psychological Design Considerations

Identity in Crisis: Customization opportunity through personalized fabric/screens, traditional cultural motif integration, sense of ownership from self-building, community formation through shared construction.

Comfort in Confinement: Hexagon reduces claustrophobia vs. rectangle, multiple facade angles provide varied views, natural ventilation ensures better air quality, community visibility reduces isolation.

Logistics & Implementation

Transport: Flat-pack design, all components fit in conventional truck, lightweight (manual handling), many units per truck.

Assembly: Simple tools (hammer, screwdriver), pictorial instruction manual, 2-4 people can assemble 1 unit in 4-6 hours.

Distribution: Government/NGO procurement, distribution through relief networks, site selection by local authorities, residents self-assemble with supervision.

Tools & Technologies

Prefabricated Timber Systems, Hexagonal Geometry, Modular Construction, Sustainable Materials, Flat-Pack Design, Community-Driven Assembly, Pandemic Response Architecture

Overview

A collection of Python scripts generating animated video sequences for a 228×8 pixel LED strip display. The show details the cycle of the four seasons: from summer bee pollination through fall color transitions, winter aurora borealis, and spring cherry blossoms — looping back to summer. Created by William Shaw, Suyeon Cha, Ananya Shrimali, and Jessica Han (Team B(ee)) for Pausch Bridge Lighting Programming at Carnegie Mellon University School of Computer Science.

1. Summer Hive Simulation

A biological simulation of bees pollinating flowers across a strip garden. 4 bee agents originate from a brown hive zone (left 12 pixels) and navigate a 216-pixel garden divided into 57 chunks. Bees select parent flowers using proximity-weighted probability, visit both parents, then plant new flowers in empty chunks. Movement runs at 0.3 pixels/frame with 15-frame hover delays, continuing until the entire strip is filled.

2. Bridge Wave Effects

Procedural wave animations using multiple sine wave layers with different frequencies, phases, and colors. Includes magenta and teal waves with snow-white sparkle accents. Features 100-frame fade-in/fade-out transitions across 1000 total frames.

3. Seasonal Transition

A 41-second HSV-based color interpolation across 100 LEDs (25 "bees" + 75 "flowers"). Summer-to-Fall transitions over 23 seconds, Fall-to-Winter over the remaining 18 seconds. Uses cosine-based ease-in-out timing and per-LED color assignment with randomized flower palette selection for smooth, natural-feeling transitions.

Tools & Technologies

Python, NumPy, OpenCV, HSV Color Space, Generative Animation, LED Strip Hardware

Overview

An Arduino-based control system for parametric window shutters designed to respond dynamically to daylight. Created for IntelEnv 2025 by team members Katie, Evelyn, and Ananya. The system targets corporate office environments where users maintain static positions throughout the day, eliminating manual blind operation while enhancing building aesthetics.

Technical Implementation

The system controls 16 servo motors (one per window pane) via an Arduino Mega, with servos on pins 3, 5, 7, 9, 11, 13, 15, 17, 19, 21, 23, 25, 27, 29, 31, 33. Serial communication runs at 9600 baud rate with smooth servo movement (15ms delay between degree increments).

Code Evolution

Version 1: Basic 4-servo control with automatic return to closed position after a delay.

Version 2: Manual control for 4 servos — type '1'-'4' to open, 'r' to reset all.

Version 3: String command system — commands like "1o" (open) and "1c" (close). Scaled to 16 servos in V3b.

Version 4 (Final): Array-based architecture for scalability, smooth servo movement with gradual angle transitions, simplified command parsing. Commands: "1o" through "16o" to open (90°), "1c" through "16c" to close (0°).

Design Philosophy

The team balanced automation, aesthetics, and practical problem-solving. Features like multi-color light filtering, origami structures, and pattern-based movement were identified for future iterations. The focus was on a robust, working prototype that demonstrates the core concept of daylight-responsive architecture.

Hardware Reference

Based on SG90 Servo Motor with Arduino . External power supply recommended for more than 6 servos.

Tools & Technologies

Arduino Mega, SG90 Servo Motors, C++ (Arduino), Serial Communication, Parametric Design

Project Overview

Type: Independent Research — Computational Neuroscience + AI
Duration: March 2026
Scale: ~3,300 lines across 5 source files + 937-line cognitive architecture spec
Model: ~300K parameter PyTorch neural network (CPU-trainable)

Concept

A digital twin system that computationally models an individual's cognitive architecture using a biologically-grounded neural network. Unlike conventional AI persona systems that replicate surface-level speech patterns, PCAE models the underlying neural machinery that produces thought, emotion, and behavior. The system functions simultaneously as a communication translator, cognitive diagnostic tool, and self-awareness instrument.

Neural Brain Model

The PyTorch model mirrors biological neural signal propagation through 8 processing stages. Input text is embedded via all-MiniLM-L6-v2 (384 dimensions), then processed sequentially through 13 specialized brain region modules: Thalamus, Amygdala, Hypothalamus, Temporal & Parietal cortex, Prefrontal Cortex (4 sub-regions), Hippocampus, DMN, Insula, Mirror Neurons, Basal Ganglia, Cerebellum, and Language Center. Each forward pass outputs a BrainState containing region activations, NT levels, and cognitive readouts (valence, arousal, stress, speech readiness, dissociation, simulation load).

Training

300 epochs using AdamW with cosine annealing. Loss combines four components: region activation BCE (1.0), neuroscience wiring constraints enforcing 7 biological rules (0.5), NT baseline anchoring (0.3), and contrastive brain vector clustering (0.2). The architecture is neuroscience-defined; individual parameterization via cognitive profile data sets gains, thresholds, and biases — the same model class can represent any person.

Frontend

Dark-theme single-page chat interface with a live brain state visualization showing 13 color-coded region activation bars and 6 cognitive readout gauges (valence, arousal, speech readiness, action readiness, dissociation, dlPFC load). Sidebar contains audience tier selector, truth level slider, and memory input panel.

Tools & Technologies

PyTorch, ChromaDB, all-MiniLM-L6-v2, Google Gemini 1.5 Pro, FastAPI, Uvicorn, Vanilla HTML/CSS/JS

Project Overview

Type: Mobile Application — 15113 Course Project
Duration: Spring 2026
Platform: Cross-platform (iOS & Android) via Expo
Repository: GitHub

Motivation

While preparing for the LEED Green Associate certification, I wanted a focused, interactive flashcard system tailored to LEED GA mock exam questions — something more effective than generic study tools. This app was built to combine exam prep with mobile development practice.

Demo Video

App Screens

Study Screen (Home Tab): The main study interface where you flip through flashcards one at a time. Tap a card to reveal the answer, use arrow buttons to navigate, and mark cards as Known or Still Learning. A filter toggle lets you focus on unmastered cards only.

Manage Screen (Second Tab): Displays all 100 flashcards in a scrollable list with colored badges indicating mastery status. Shows overall progress at the top with an option to reset all progress.

Technical Implementation

Built with React Native and Expo SDK 54, using file-based routing via expo-router. The card flip animation leverages a Y-axis rotation built with react-native-reanimated for a native-feel interaction. All 100 questions are stored in a typed data file (data/questions.ts), and card progress is persisted locally through @react-native-async-storage/async-storage. The app uses a consistent light theme with green (#4CAF50) and orange (#FF9800) accents.

Process & Takeaway

After learning the value of thorough planning in a previous assignment, I invested heavily in writing a clear spec and workflow plan before writing any code. The structured approach paid off — the core app was built in a single agent-assisted session with only minor tweaks needed (adding answer options to card faces). The engineering architecture required no revisions.

Tools & Technologies

React Native, Expo SDK 54, TypeScript, expo-router, react-native-reanimated, AsyncStorage, Expo Go

Project Overview

Type: Web Application — 15113 Capstone
Duration: Spring 2026
Stack: Next.js 16 (App Router), React 19, TypeScript, Tailwind CSS 4
AI: Google Gemini Flash (gemini-flash-latest) via @google/generative-ai
Repository: GitHub
Live App: vibecheck-15113-capstone.onrender.com

Elevator Pitch

VibeCheck recommends music from implicit context instead of explicit mood surveys. It collects passive signals in the browser — time, timezone, optional weather, device hints, idle estimate, and on-device listening history — sends them to a server-side LLM that returns a strict JSON "vibe card," then surfaces three playable tracks via the iTunes Search API. No accounts. No mood checkboxes. Just open and play.

Demo

Motivation

Music discovery often means manual search or static playlists. VibeCheck experiments with: if we know something about the user's moment — clock, environment, recent listens — can we shorten the path from "open app" to "good song playing"? The project also demonstrates responsible AI use: structured outputs, fallbacks when the API is unavailable, and non-clinical copy.

Architecture

Two server-side API routes handle all AI and music logic. The browser calls /api/infer with a SignalsPacket — local time, optional weather, device hints, idle estimate, recent listening history, and an optional vibe profile — and Gemini returns a typed InferenceResult JSON. That result's deezer_search_query field feeds /api/music, which merges paraphrased iTunes search queries, picks three diverse tracks by artist/genre/decade/album, and optionally resolves full-length audio via public Piped mirrors — no YouTube Data API key required. No database; all client state lives in localStorage.

Privacy & Security

Consent screen before any signal collection. Gemini API key lives server-side only (process.env.GEMINI_API_KEY); .env* files gitignored. No OAuth, no accounts — all listening history and preferences stay in localStorage with user-accessible clear controls in the footer. Geolocation stored locally for repeat weather fetches; user can skip location entirely and the app still functions.

Limitations & Future Work

iTunes and Piped availability varies by region and network. Gemini quality and cost depend on API rate limits. Mobile is responsive web, not a native app. Possible next steps: native mobile or PWA + push notifications, optional last.fm / Spotify read-only OAuth, embeddings-based catalog search for tighter mood–track alignment, and a stronger accessibility audit.

Tools & Technologies

Next.js 16 (App Router), React 19, TypeScript, Tailwind CSS 4, Google Gemini Flash, Apple iTunes Search API, Open-Meteo (weather), Piped (full playback), localStorage (multi-module client persistence)