I build products, and I study the mathematics underneath them.
I'm a computer science engineer from Punjab, India. I co-founded FemoraAI, a women's health app used by more than 4,600 people, and now Luna, an AI executive assistant for founders. I work on software that has to make sense of messy, incomplete information, and on the numerical methods and GPU code that make it fast and trustworthy.
Website: ramanjitsingh-hub.github.io
Each investigation is a question, a derivation, code written from scratch, and measured results, including the ones that did not go my way.
| 01 | Re-deriving a production cycle predictor | The Kalman filter I shipped at FemoraAI, ported line by line. It becomes a moving average after five cycles, and a 3σ safety gate can lock it out permanently. An exact Bayesian replacement fixes both. Synthetic data. |
| 02 | A Navier–Stokes solver on a 4 GB laptop GPU | CUDA from scratch. Matches an independent FFT solver to 5.6 × 10⁻¹⁶, exactly second order, validated on the lid-driven cavity, then 13.5× faster with the same answer. It also runs in your browser on WebGPU. |
Code for both: ramanjitsingh-hub.github.io
- Luna (2026 –), co-founder. Agent architecture, AI infrastructure, backend.
- FemoraAI (2025 – 2026), co-founder. Backend, AI inference, data infrastructure. Per-user Kalman filtering, 7 microservices, a team of 21.
- Pink Parcel (2023 – 2025), co-founder. Period kits to six university hostels in seven minutes.
Working with: CUDA · C++ · Python · NumPy/SciPy · PyTorch · Java/Spring Boot · TypeScript · WebGPU · MongoDB · Kafka · GCP/AWS