Shipped features and performance improvements across open source projects - from ML emissions tooling to civic health infrastructure for low-bandwidth environments.

• mlco2 / codecarbon

  ML carbon emissions tracking

  ~1.9k starsPython

  Merged#1219 / #1235· merged into v3.2.8 + master· Jun 7 & 11, 2026

  CodeCarbon performance optimizations

  Profiled CodeCarbon's hot paths and shipped two rounds of optimizations — reducing unnecessary work in the measurement loop, file I/O, and CSV handling.

  • #1219 — Eliminated unnecessary list allocation on every measurement tick
  • #1235 — CSV header validation now reads one row instead of the entire file

  Proposal#1237

  Performance improvement: Avoid unnecessary GPU detail queries in _monitor_power()

  The monitoring loop was querying detailed GPU info (memory, temperature, processes) every second — but only utilization was needed. Proposed a lightweight helper that skips the expensive calls.

  • Avoids querying GPU memory, temperature, and process lists on every tick
  • Saves dozens of NVML calls per second on multi-GPU systems

  Pythonpsutilpandasthreading
• RatLoopz / sahidawa-india

  Civic health platform · GSSoC 2026

  ~55 starsNext.js

  Merged#1021· merged into main· Jun 1, 2026

  PWA · Offline-first service worker caching for medicine lookup

  Built a hand-rolled offline-first PWA caching layer for SahiDawa — a platform helping citizens in India verify medicines and find trusted pharmacies in low-bandwidth environments. Designed distinct caching strategies per route type and added 8 tests covering all strategies and SW lifecycle. Labeled advanced · quality: exceptional by maintainers.

  • StaleWhileRevalidate — medicine APIs
  • CacheFirst — OSM tiles (7d TTL)
  • NetworkFirst — alerts
  • 143/143 tests passing

  Next.jsService WorkersPWAOpenStreetMapTypeScriptJest

1. 

   Software Engineer · CrowdDoing

   San Francisco, CA · Part Time

   May 2026 — Present

   • Architected an event-driven MCP agent in Python that matches 1,000+ weekly volunteer profiles to technical roles, cutting onboarding from 24 days to 6 days, by integrating Open WebUI for conversational access and transforming 5,000+ legacy records into retrieval-optimized datasets.
   • Implemented JWT-based authentication and role-based authorization middleware enforcing role-scoped access to volunteer PII across the MCP and Open WebUI layer, preventing unauthorized access to 40k+ volunteer records, by adding per-request audit trails and role-based access controls.

   PythonAgentic AILLM
2. 

   Software Engineer (Graduate Consultant) · Nissha Medical Technologies

   Buffalo, NY ·

   August 2025 — December 2025

   • Automated proactive quality analysis for ~1M daily manufacturing units

   • Led a team of 4 engineers to deliver a computer-vision inspection system that automated quality analysis for ~1M daily manufacturing units, earning pilot-deployment approval, by gathering requirements with stakeholders and designing the solution methodology.
   • Owned the full ML lifecycle from data annotation through YOLOv8 training, achieving 98.7% mAP50 on the primary defect class and 79.3% overall across 7,253 instances, by fine-tuning models and building an OpenCV preprocessing and edge-inference pipeline for sub-50ms latency.

   PythonOpenCVYOLO
3. 

   Software Developer Engineer · Siemens

   India · Full Time

   June 2019 — July 2024

   • Production services at industrial scale
   • Observability & reliable system design

   • Engineered C#/.NET Core and Python backend services processing 100K+ daily energy events on the Building-X platform, achieving exactly-once processing semantics via idempotency-key-based deduplication, by designing scalable event-driven microservices and APIs.
   • Built event-driven AWS Lambda and DynamoDB microservices in Go backed by DynamoDB for energy demand forecasting and real-time consumption visualization, leading to 30% faster forecast refresh for building operators, by processing multi-resolution timeseries across distributed EventBridge and SQS/SNS pub/sub patterns.
   • Developed 3 Angular and TypeScript micro-frontend dashboards that gave building operators real-time visibility into consumption, carbon-emission baselines, and demand forecasts, by rendering 150K+ energy KPI datapoints through a Node.js BFF.
   • Containerized Flyway SQL migration scripts with Docker and migrated CI/CD pipelines from Jenkins to GitLab, cutting environment setup time 80% and deployment time 50%, by adding feature toggles and Grafana and CloudWatch alerting.
   • Won Product Category at Hackathon-19 by prototyping a proactive maintenance dashboard enabling field engineers to monitor and manage access controls in real time.
   • Reduced onboarding time from 3 weeks to 10 days by mentoring 3 junior engineers and onboarding 5 new team members through structured code reviews, pair programming and technical documentation.

   C#.NET CoreGoAWSMySQLDynamoDBCI/CD
4. 

   Software Engineer Intern · Siemens

   India · Full Time

   January 2019 — June 2019

   • API optimization
   • Quality & reliability

   • Implemented NUnit testing frameworks in .NET Core, raising code coverage to 93% and SonarQube quality to Grade A and establishing the testing standard adopted by the full team, by writing automated unit and integration test suites.
   • Optimized request and response handling for key workflows, reducing REST API latency 40%, by adding query optimization and response caching.
   • Designed recursive tree-based visualization prototypes in Angular, adopted as the core visualization structure in Siemens Navigator's DDA tool, by modeling buildings, sites, and meters as hierarchical nodes.

   .NET CoreNUnitREST APIsAngular

• 20263 min read

  From Research Paper to Production: Why AI Systems Need More Than Models

  Model is one part of the system

  Read on Medium
• 202612 min read

  Your LLM Judge is Lying to You (And Here’s How to Fix It)

  Evaluation biases and suggested mitigations

  Read on Medium
• 20265 min read

  From Software Engineering to AI

  Transitioning from industrial backend development to graduate study in artificial intelligence.

  Read on Medium