|
Pratham Gandhi
I build assistive computing and safe autonomy. My work spans haptic‑audio interfaces for non‑visual programming (HACI), contract‑based design with Signal Temporal Logic (COASTL), and accessible no‑code ML tools (AnaLazy).
I’m a data scientist at Dell Technologies, working on human‑centered AI for global sales planning.
Email /
CV /
Google Scholar /
GitHub
|
|
Work
♿ Accessibility & HRI
🎓 EdTech
|
HACI: A Haptic‑Audio Code Interface for Non‑Visual Programming
∞
Honors Thesis, University of Chicago, 2024
A browser IDE + Arduino glove that combines structured speech with finger‑localized haptics to reduce cognitive load in non‑visual code navigation and debugging.
Runtime: Ace Editor with AST‑aware guidance (Acorn), Web Speech + non‑speech cues, Web Serial to an Arduino Mega driving six DRV2605L vibrotactors via I²C multiplexer.
Pilot highlights: improved structural wayfinding and clear next steps around cue consistency and customization.
paper /
code /
slides
|
🛡️ Safe Autonomy
∑ Formal Methods
|
COASTL: Contract Operations & Signal Temporal Logic for Safe CPS
∞
USC DesCyPhy Lab — Python library
A design‑by‑contract toolkit that parses STL into syntax trees and composes assumptions/guarantees to enforce requirements early in CPS design.
Encoding: prototype translation from STL specs to mixed‑integer constraints for behavior synthesis; supports illustrative examples and poster artifacts.
poster /
code
|
🧰 Data Tools
⚙️ No‑Code ML
|
AnaLazy: No‑Code ML with Built‑In Data Quality
∞
GUI app for spreadsheet‑to‑model workflows
A graphical app that ingests CSV/XLS data, runs quick data‑quality checks, then guides feature/target selection and model training (classification/regression).
Emphasis on responsible defaults: quality checks first, clear evaluation views, approachable UX for non‑programmers.
Flow demo: import → checks → select → train → evaluate → export.
write‑up /
demo /
code
|
Earlier work (Toksik, Space Walk, TIPP&SEE, SSI)
🗣️ Responsible AI & NLP
|
Toksik: Classifying Harmful Language in Social Media
∞
Supervised NLP baseline (NB vs. LR) with error analysis
Transparent pipeline with feature engineering, confusion matrices, and thresholding; Naive Bayes ~92% vs. Logistic Regression ~87% on held‑out data.
paper /
code
|
🎮 Learning Game
|
Space Walk: A Game Metaphor for Quantum Entanglement
∞
Two‑player collaborative web game for conceptual understanding
Coupled mechanics mirror entanglement’s state correlations; formative prototyping and play‑testing.
report /
code
|
🎓 Learning Sciences
|
How Middle‑Schoolers Use TIPP&SEE in CS
∞
Unpublished manuscript (4th–7th grade; Scratch; metacognition)
Behavior analysis across TIPP&SEE phases; implications for prompt sequence and scaffolding.
manuscript
|
🔗 Causal Inference
|
Education → Health: A Cross‑Country Causal Pathway
∞
SSI project: OLS per indicator country with socio‑economic controls
Education → Employment → GDP → Medical Care → Life Expectancy, with paired countries to interpret heterogeneity in the mechanism.
research design slides /
paper
|
|