Completed FlyRank's machine learning internship from July 1 to October 5, 2026: 23 applied assignments, 152+ documented hours and an accepted capstone on Google Search Ranking & Discoverability.

• Built a content-refresh ranking pipeline on 37,385 pseudonymized pages from FlyRank's production search warehouse. A random forest ranks pages already losing impressions for editorial review; the study measures existing decline, rather than forecasting later traffic.
• Caught and removed feature leakage that had inflated Precision@50 to roughly 1.0: prior-window impressions, combined with another feature, could reconstruct the decline label. Rebuilt the feature frame and recorded leave-one-client-out Precision@50 of 0.258 ± 0.237 against a same-split baseline of 0.173, with substantial variation between clients.
• Designed three validation regimes: a six-client holdout, leave-one-client-out across 28 clients, and February-to-March stability checks. Reported each ranking separately to expose the gap between pooled holdout performance and cross-client generalization.
• Constructed a CTR-headroom baseline, then recalibrated it against the observed signal distribution after an initial version collapsed into ranking by impressions. Evaluated the baseline and learned model on identical splits.
• Designed a proposed seven-archetype editorial playbook, including high-leverage refresh, snippet optimization, content aging and stable-performer protection. Added fixed reason codes, human-review rules, editor overrides and explicit limits on automation; the proposed actions were not validated against later traffic.
• Built a scheduled research-scout agent with Python, the arXiv API, GitHub Actions and a pytest evaluation harness. It produces a weekly ranked digest of information-retrieval and ML literature, with a relevance gate and a strict source allowlist.
• Published the full pipeline as a research paper on GitHub Pages, covering the data contract, baseline, model, leakage audit and action playbook. Kept claims traceable to committed notebooks, saved metric receipts and regenerable figures.

Read the research paper ↗

Tools Used: PostgreSQL, Javascript, Node.js, Express, Redis, Prisma, Socket.IO, Azure

Worked as a Backend Engineer Intern at Docwo Pvt. Ltd. from April 2026 to June 2026 and worked with a team of 9 members throughout this time period.

Architected, Built and deployed the entire backend service of an ambulance dispatch app (ResQ).

Supports patient, driver, and admin applications with real‑time ambulance discovery, booking, driver assignment, and live GPS.

Hardened the platform with JWT authentication (refresh token rotation), Google OAuth 2.0 integration, role‑based access control, rate limiting on critical endpoints, and AES 256 GCM encryption for PII fields and progressive OTP lockout with attacker detection.

Comprehensive admin dashboard with real time KPIs, growth reports, support tickets, and full CRUD; engineered automatic driver matching (proximity search, accept/reject with timeout retry, escalation), Socket.IO live tracking, transactional notifications, Swagger API documentation, and production deployment on Azure App Service.

Won the Best App award in the final weekly sprint !