Work that solves problems.

CONTEXT: Built to solve the chaos of managing a 100-bed student hostel that relied on paper ledgers and physical keys. CONSTRAINTS: Zero budget for infrastructure. Needed to run on a cheap cloud instance. Users (staff) were non-technical and resistant to "new tech." DECISIONS: Opted for a "mobile-first but desktop-capable" PWA. Used a Postgres backend for strict data integrity over flashy NoSQL. RESULTS: Reduced check-in time from 15 mins to 90 seconds. Eradicated "double-booking" errors which were costing ~$200/mo in lost revenue. CURRENT STATE: Maintenance mode. The core logic hasn't needed a rewrite in 2 years, proving that boring tech wins. NEXT TIME: I'd use a more robust state management library from day one. Prop-drilling became a nightmare as the feature set grew.

CONTEXT: A real estate play. Took a distressed property and converted it into a premium student living facility. CONSTRAINTS: Tight margins. Every dollar spent on "luxury" had to translate to a 1.5x increase in rental yield within 6 months. DECISIONS: Focused purely on "The Three Pillars": High-speed internet, reliable power, and clean water. I cut the budget for common-room TVs and redirected those funds into premium orthopedic mattresses. RESULTS: Achieved 100% occupancy within 3 weeks of launch. 20% higher rent compared to neighboring properties with "better" visual aesthetics. CURRENT STATE: Cash-flowing and stable. Management has been delegated to a site lead using the SaaS mentioned above. NEXT TIME: Hire a professional contractor earlier. Doing the initial electrical planning myself was a "founder trap" that wasted two weeks of my time.

CONTEXT: A tool to extract table data from 10,000+ unstructured logistics PDFs monthly. CONSTRAINTS: Data accuracy had to be >98%. Hand-keying was the "baseline" to beat in terms of cost. DECISIONS: Combined Amazon Textract with a custom Python post-processing layer. I chose not to build a custom LLM; instead, I built a rigid regex-based validation engine to catch Textract's hallucinations. RESULTS: Automated 85% of the workflow. The client saved ~40 man-hours per week, allowing them to scale without hiring more back-office staff. CURRENT STATE: Successfully handed off to the client's internal IT team. NEXT TIME: I should have spent more time on the error UI. When the system failed, it didn't tell the user *why*, leading to unnecessary support tickets.

CONTEXT: A peer-to-peer carpooling app for high-density tech parks. CONSTRAINTS: High trust barrier. People are wary of strangers, even coworkers. DECISIONS: Restricted sign-ups to verified @corporate emails only. I focused on "Micro-Transactions" to handle fuel sharing without making it feel like a commercial service. RESULTS: Validated the demand with 50+ user interviews. However, the unit economics didn't make sense without a massive marketing spend. CURRENT STATE: Shelved. It was a great technical exercise but a poor business model for a solo founder. NEXT TIME: Validate the distribution channel *before* writing the first line of code. The "Build it and they will come" approach failed here.