20 · Dropout · Builder — Bridging AI research and production engineering.
I dropped out of NIT Kurukshetra (3rd year) to co-found AuraX, a fashion-tech startup funded by IIIT-Hyderabad's CIE. We built on top of diffusion models that beat Google, Stable Diffusion, most competitors in virtual try-on quality, signed deals with Aditya Birla Fashion Group and worked with brands from Mensa Brands, so on..., and closed when bigger players flooded the space. Before that, I built SynthAiLabs (OpenEdu) — a learning community where juniors restarted the mission on their own after I moved on.
Now I build products that scale.
| Project | Description | Stack | Link |
|---|---|---|---|
| CCProxy | A universal proxy to supercharge Claude Code with any AI provider—OpenAI, Gemini, or local Ollama models. | TypeScript, AI Proxy | Website · GitHub |
| OpenCrow | Open-source AI agent platform — embeddable chat widget with dual-layer tooling (server + client) and dynamic tool merging for LLMs | React, TypeScript, Node.js, OpenAPI | Website · GitHub |
| AI Inference Platform | Smart-routing inference with query classification, multi-level caching (L1 Moka + L2 Redis), semantic matching, and full Prometheus/Grafana observability | Rust, Axum, Tokio, Redis, Docker | Website |
| TuneSLM | Drop a dataset, answer a few questions → get a Colab notebook to fine-tune your SLM via Unsloth | Python, Unsloth, Google Colab | Website |
| MockServer | CLI tool for mock backends with a GUI — adjustable latency, error injection, instant frontend dev | TypeScript, Next.js, Express | Website · GitHub |
- Conflict-Aware Adapter Composition (C-AAC) — Advanced scene generation via conflict-aware adapter merging (AuraX / IIIT-H CIE)
- ReNU: Texture & Color Fidelity in Diffusion Models — Outperforms baselines on SSIM, FID, and CIEDE2000 (AuraX / IIIT-H CIE)
- MoE: Brain-Inspired LLM Architecture — Whitepaper on MoE as a computational analogue to functional specialization (SynthAiLabs)
- GPT-Neo — Full implementation of "Attention is All You Need" from scratch in PyTorch
Interests: Diffusion Models, LLM Architectures, Distributed Systems, Quantum Computing, Compiler Design
- Languages: Python, TypeScript, Rust, Go, C++, Swift
- Frameworks: Next.js, Nest.js, React, Prisma, Tauri, Tailwind, PyTorch, TensorFlow
- AI/ML: Diffusers, LLaMA, Unsloth, Scikit-learn, CUDA, Qiskit
- DevOps: Docker, AWS, ShaktiCloud, GCP, CI/CD, GitHub Actions, Linux
- Databases: PostgreSQL, MongoDB, Redis, TimescaleDB, Milvus
More at himanshuat.com