Jatin Besoya

Owned end-to-end development of a custom in-house analytics platform for a tech media site serving 2M+ monthly sessions. Built a rich front-end capture layer tracking scroll depth, retention time, page-to-page navigation, input interactions, and click events with full contextual metadata — including which option was selected within dynamic components (carousels, tables) and what peer options were visible at the time of interaction. Engineered the full data path: SQS ingestion, MongoDB storage, ClickHouse migration pipeline, and insight delivery. Queried and delivered behavioral insights to product, design, and leadership — directly influenced layout decisions, section ordering, content prioritization, and surfaced cross-session patterns including return frequency, full user interaction timelines, and funnel drop-offs.

Designed and implemented a custom query-scoped in-memory caching layer in Go for a high-traffic tech media platform. Each cache instance is isolated from specific query rules and self-terminates due to staleness or updates, eliminating manual cache invalidation. Built with full MongoDB API compatibility, the system required zero application-side changes to integrate and significantly reduced primary database load on heavy read paths. The caching layer eventually replaced Redis for specific workloads, contributing to an overall ~80% reduction in primary database load while ensuring back-end performance and data consistency.

Founded the technical foundation of a pre-sold B2B coaching platform from zero at OriginBluy — deleted an unscalable prior back end, redesigned the full data model in MongoDB, set the React/Next.js folder and file architecture, and trained the team to follow it. Built a feedback analysis system with unique-link feedback forms for peers/seniors/juniors/managers, 0–10 competency scoring, before/after coaching graphs, and AWS SES for transactional email delivery. Built organization manager dashboards with pre-computed MongoDB aggregations refreshed at write time, eliminating expensive on-read aggregation. Built Jest + Supertest automation that cut bug-replication time from ~2 hours to under 6 minutes, surfacing hidden bugs in the assessment pipeline.

Architected and shipped an AI-powered video assessment platform end-to-end at OriginBluy, integrating AI-powered analysis of 32 behavioral parameters (confidence, gestures, pauses, filler words, eye contact, volume). Built front-end routes, AWS EC2 ingestion, custom MongoDB-backed ticket queue, and a hybrid cloud architecture (AWS EC2 + on-prem Linux GPU server with RTX 4090) — the on-prem choice was a deliberate cost decision since the same GPU on cloud was significantly more expensive during the MVP phase. Engineered an async pipeline with FFmpeg for WebM-to-MP4 conversion, Python/Flask inference, Node.js cron for status sync, and EJS + Puppeteer for PDF report generation. The custom queueing system included local server health monitoring, status tracking, and fallback state. Used for both employee coaching assessments and hiring candidate interviews.