Pradeep Periyasamy

Texas A&M University | College Station, TX

• Architected Flask → FastAPI migration for Texas Congestion Analysis Tool (TCAT), modernizing statewide transportation analytics platform serving TxDOT planners and engineers; designed incremental rollout with shared business logic layer enabling 3–5× concurrent request handling without downtime.
• Engineered legacy → React frontend modernization (component-based architecture, Redux state management, React Query for server-state caching); established component library (map visualization, tables, report forms) enabling parallel frontend/backend development.
• Built National Bridge Structural Failure Engine analyzing 624K+ bridges across 28 years of historical NBI data using DuckDB + Parquet (columnar storage, projection pushdown) enabling multi-gigabyte pipeline on 16GB RAM without expensive infrastructure scaling.
• Engineered XGBoost-based risk prediction system with separate models for 7 failure categories (scour, collision, deterioration, etc.); handled class imbalance via oversampling + threshold tuning + cost-weighted loss, improving rare-event detection (collision failures ~1% of data).
• Implemented T−1 temporal labeling pipeline matching historical NBI snapshots to documented bridge failures, eliminating survivor bias; validated against state-level records and confidence thresholds (year-built consistency, anomaly detection) reducing false-positive labels by 40%.
• Designed SHAP-based feature importance analysis deriving global and local explanations for each bridge's failure risk; surfaced top-5 risk drivers per failure category (e.g., waterway conditions dominate scour predictions, clearance dominates collision predictions).
• Validated models using ROC-AUC, PR-AUC, and calibration metrics on held-out 2020–2025 failures; tested for geographic bias (rural vs. urban Texas) and age bias (new vs. old bridges) ensuring fair scoring across diverse bridge populations.
• Generated 25K-bridge shortlist (4% of inventory) prioritizing inspection/maintenance based on risk thresholds (scour >60%, collision >50%, deterioration >70%); directly informed TxDOT's accelerated inspection roadmap and maintenance budget allocation.

Indian Institute of Technology Delhi | Delhi, India

• Architected and shipped B-Reporter web platform (React.js, Next.js, MySQL) enabling university-wide collaborative reporting with commenting, tag-based post suggestions, and responsive UI — deployed across IIT Delhi.
• Designed tag-based recommendation system leveraging user engagement patterns to surface relevant posts without explicit search, improving content discoverability and user retention.
• Implemented real-time commenting system with nested replies, user mentions, and activity feeds using React state management and MySQL transactions for consistency.
• Built responsive UI with mobile-first design (CSS Grid, Flexbox) ensuring accessibility across 375px–1920px viewports; optimized load time to <2 sec on 3G networks.
• Engineered data pipeline normalizing user-generated content and metadata into relational schema, supporting 1000+ concurrent users without performance degradation.
• Led frontend architecture decisions (component composition, state management patterns) establishing best practices adopted by subsequent development batches.

Optmyzr | Hyderabad, India

• Architected .NET data-caching service reducing redundant Bing Ads API calls by 97%, eliminating per-team duplicate fetch logic and unifying data retrieval across platform.
• Integrated 43 Bing Ads widgets via async pipelines, achieving 95% feature parity with existing Google Ads platform while maintaining zero performance regression on dashboard load times.
• Engineered async/await patterns handling millions of daily widget impressions without database saturation; optimized connection pooling and batch processing for throughput at scale.
• Designed data persistence layer with intelligent TTL-based cache invalidation, ensuring stale data didn't block new API updates while minimizing redundant API round-trips.
• Implemented comprehensive integration tests covering edge cases (API timeouts, partial failures, concurrent requests) ensuring 99.5% uptime during production launch.
• Shipped feature within 10-week internship sprint with zero critical incidents; results directly influenced platform's multi-platform architecture roadmap for subsequent years.

Infigon Futures | Mumbai, India

• Built full-stack website from scratch using React, TypeScript, Redux, Firebase, and Docker; deployed containerized application achieving <1 sec load time and mobile-responsive design across all devices.
• Designed Redux state management architecture handling complex user workflows, enabling predictable state transitions and maintainable component composition across 15+ pages.
• Integrated Firebase authentication, real-time database, and hosting, implementing OAuth login flows and Firestore data synchronization supporting concurrent user sessions.
• Led 4-member web development team through 8-week sprint; conducted code reviews ensuring TypeScript best practices, organized KT sessions on component patterns, and unblocked team members on technical blockers.
• Established Git workflow (feature branches, squash commits, CI pre-checks) and linting standards using ESLint/Prettier, improving code quality and reducing merge conflicts.
• Shipped production website with zero critical bugs; stakeholders adopted platform for internal operations, validating architecture and UX decisions established during internship.

Indian Institute of Technology Delhi | Delhi, India

• Architected and shipped B-Reporter Android application from scratch using Kotlin, MVVM architecture, and Node.js backend; achieved 95%+ test coverage across business logic and successfully launched on Play Store.
• Implemented MVVM pattern with Android Architecture Components (ViewModel, LiveData, Room database) ensuring separation of concerns, testability, and lifecycle-aware state management.
• Built real-time communication layer using REST APIs, handling network errors gracefully with retry logic and offline-first caching via Room database for seamless user experience.
• Optimized APK size to 8.5MB through ProGuard obfuscation and dependency management; achieved 60 FPS scroll performance on mid-range devices (4GB RAM) via efficient UI rendering.
• Designed Node.js backend endpoints (Express.js) supporting authentication, CRUD operations on user-generated content, and real-time data synchronization with Android frontend.
• Shipped production app within 4-week sprint with zero critical bugs; Play Store launch validated full-stack mobile architecture and drove 100+ installs in first month.

Delta Force (Software Development Club) | NIT Tiruchirappalli

• Architected Kaval Arann — public-police communication & incident reporting platform deployed in production across Tiruchirappalli City Police; dual-app system (citizen-facing incident reporting app + police-facing incident management dashboard) in active use.
• Engineered single-codebase, multi-flavor Android architecture generating two distinct APKs from shared codebase (Retrofit API client, Room data layer, MVVM business logic); achieved 40% reduction in duplicate development effort and maintenance burden.
• Designed navigation architecture using Android Jetpack Navigation Component (15+ screens, type-safe argument passing via Safe Args, shared ViewModels) enabling seamless transitions with near-zero navigation-related crashes.
• Implemented offline-first persistence layer using Room with repository pattern; local drafts sync automatically via Retrofit REST API once network resumes — critical for low-connectivity civic reporting scenarios.
• Built end-to-end incident reporting pipeline: structured data capture, FusedLocationProvider geolocation, CameraX-based media evidence (image/video with watermarking), real-time police dashboard for triage and status management.
• Designed role-based access control at API layer (Retrofit interceptors) and UI layer (role-gated fragments) supporting citizen submissions and police operational workflows; enforced principle of least privilege across 10-member team.
• Mentored 24 incoming Delta Force developers on scalable architectures, Git collaboration, code review culture, and debugging practices; established curriculum adopted in subsequent cohorts and promoted 5 junior developers to leadership roles.
• Led FunLear v2 open-source project as part of Delta Winter of Code, mentoring first-time contributors on codebase navigation, branching strategies, constructive code review, and transition from individual programming to collaborative software engineering.