Software Development Engineer II @ SpeEdLabs
AI and backend engineer building production LLM systems and the Python services that hold them up. Built for exam day, not for demos. At SpeEdLabs (SDE II) I own the LLM orchestration layer across OpenAI, Anthropic and Gemini, where batch-first routing cut inference cost 40%, and the OMR grading pipeline behind 225K students' exams. Before that I built a computer-vision diagnostic system and validated it at AIIMS at 93% accuracy. My sweet spot: reliability, cross-store consistency, and applied AI that survives production.
Experience
Software Development Engineer II
SpeEdLabs · Hyderabad
Nov 2025 – Present · 11 mo
EdTech SaaS platform serving 225K students at 450+ schools. • Own the OMR evaluation pipeline for bulk sheet ingestion and automated grading (OpenCV, NumPy), using perspective transformation and fiducial-based alignment to raise scan accuracy from 80% to 98%. • Own multi-provider LLM orchestration across OpenAI, Anthropic and Gemini, cutting inference cost ~40% by routing work through batch APIs first, with inline fallback and tiered exponential-backoff retries. • Designed the pytest architecture the backend team builds on: transactional savepoint isolation, dependency overrides and fixture-leak guards, so integration tests run against live SQL Server and MongoDB without state leakage. • Built an incremental metrics warehouse with cron-scheduled, state-tracked refreshes and precomputed aggregates, moving dashboards from full-history scans to summary-table reads. • Offloaded grading, reporting and LLM workloads onto Azure Service Bus queues to keep the Functions-hosted API responsive, and routed uploads directly to Blob Storage via scoped SAS tokens. • Built a unified transaction lifecycle coordinating SQL Server and MongoDB writes in one request boundary, with rollback and compensation so a failure in either store never leaves partial state. • Designed REST APIs across assessment, practice and reporting domains with JWT auth, RBAC and Pydantic validation.
Software Development Engineer
Neuranics Pvt Ltd · Gurgaon
Jun 2023 – Oct 2025 · ~2.4 yrs
AI-based diagnostic devices • Architected a multithreaded Python instrument-control service synchronising camera, 3-axis stage motion and image capture through a thread-safe command queue, for stable unattended operation across long imaging runs. • Engineered cell-detection and morphometric analysis (OpenCV, NumPy), validated by deploying the device on-site at AIIMS and matching output against lab reports with 93% accuracy. • Designed a config-driven hardware abstraction layer so new camera backends could be added without application code changes. • Built a type-safe API bridge generating TypeScript client interfaces from Python handler signatures, keeping frontend and backend contracts in sync automatically. • Built real-time calibration and hardware-validation tooling that cut operator setup effort by roughly 70%. • Automated full-range autofocus with edge-contrast sharpness scoring, removing manual focusing from every imaging run.
Software Development Intern
SmartTraveller · Remote
Jul 2022 – Jul 2022 · 1 mo
Built a Node.js and Express recommendation service aggregating travel inventory from multiple provider APIs into MongoDB, scoring options on weighted price, duration, and rating signals for per-user ranking.
Education
Bachelor of Science Economics
Indian Institute of Technology, Kanpur
2019 – 2023
Skills & stack
Candidate trust score
No referral activity yet.
