Solving Real-World Problems Through Engineering
Discover the real-life usability, target outcomes, and technical architecture behind my core applications. I build products to solve tangible challenges, ensuring robust performance and modular engineering.
InterviewPilot — AI Interview Coach
Democratizing high-fidelity professional coaching with interactive, real-time voice & behavioral AI.
🔴 The Real-World Problem
Job seekers experience massive anxiety and lack access to realistic, high-pressure mock interviews. Traditional preparation relies on static, generic questions that fail to simulate conversational dynamics. Human coaches are highly expensive, and online portals fail to provide instant, quantitative feedback on voice inflection, behavioral cues, and technical depth.
🚀 Solution & Usability Flow
InterviewPilot bridges this gap by engineering a production-grade, conversational mock interview agent. Users upload role-specific job descriptions, and the system uses Gemini AI to generate custom, context-aware interview boards. Integrated with Vapi, it conducts ultra-low latency voice interviews over WebSockets. It records user telemetry, delivering structured analytics covering sentiment, speech rate, and technical accuracy. A candidate selects their target job category, starts a real-time voice call, and interacts directly with the AI interviewer. The agent probes deep concepts, adapts dynamically to candidate answers, and provides a quantitative dashboard detailing critical improvements.
💼 Real-Life Usability & Value
Dramatically reduces candidate interview anxiety, builds physiological muscle memory for remote technical tests, and democratizes institutional-grade preparation at zero cost.
🛠️ Architectural Decisions
Next.js provides dynamic hydration and SEO optimization. Gemini AI governs adaptive question trees and automated performance grading. Vapi coordinates real-time audio streams, and NeonDB serverless PostgreSQL secures analytic progress pipelines.
PneumoAI — Pneumonia Detection
Empowering clinical staff with high-precision computer vision triage for emergency medical imaging.
🔴 The Real-World Problem
Radiology departments in public clinics face high patient backlogs. Fatigued radiologists must manually examine hundreds of chest X-rays daily under intense pressure. A delay in triaging acute pulmonary infiltration can lead to severe clinical complications, presenting a significant clinical risk.
🚀 Solution & Usability Flow
PneumoAI integrates a custom deep Convolutional Neural Network (CNN) trained on thousands of chest X-ray images, serving as a rapid second-opinion diagnostic support. The clean portal allows emergency nurses to upload X-rays and instantly visualize probability maps highlighting consolidation, congestion, or normality. Clinical personnel upload high-resolution DICOM or standard imagery at the point of care. The backend processes the tensor array in milliseconds, immediately flagging abnormalities and sorting high-risk patient files to the top of the radiologist's manual review queue.
💼 Real-Life Usability & Value
Shrinks diagnostic triage delays from hours to milliseconds, serves as a powerful medical safety net against human fatigue, and provides automated screening support for rural health centers lacking specialist radiologists.
🛠️ Architectural Decisions
PyTorch powers the deep learning training pipeline and local inference engine. FastAPI drives highly concurrent asynchronous image ingestion endpoints, and React maps diagnostic overlays instantly onto clean medical tablet screens.
CoinPush — Crypto Screening App
Sanitizing volatile market telemetry into actionable, high-performance dashboards.
🔴 The Real-World Problem
Cryptocurrency markets operate continuously, producing an overwhelming torrent of volatile price feeds and trading volumes. Casual traders and investors are flooded with noise, making it extremely difficult to identify breakout categories or trending coins without purchasing expensive institutional terminal licenses.
🚀 Solution & Usability Flow
CoinPush aggregates and structures raw market telemetry into an elegant, high-throughput screening terminal. It consolidates volatile feeds into curated trends, gainers/losers screens, and historical data, making it easy to identify genuine market-moving momentum instantly. Traders monitor live cryptocurrency movements sorted by volume velocity and trading groups. Users filter global assets by customized parameters and access technical price trends in real-time without refreshing the page.
💼 Real-Life Usability & Value
Saves retail traders hours of manual scanning across fragmented platforms, eliminates expensive data subscription fees, and delivers an intuitive portfolio-monitoring experience.
🛠️ Architectural Decisions
Next.js drives static layout caching with dynamic hydration. CoinGecko REST APIs supply high-signal price and historical data, and SWC compilation minimizes interface latency to deliver responsive page interactions.