B.Tech CSE · KL University Hyderabad, Class of 2028 · ML + systems
I build things that have to hold up under scrutiny — screening models that report their confidence intervals, spend policies that can prove their own audit log wasn't edited, and tools I use every day on my own machine.
| Project | What it does | Stack |
|---|---|---|
| omarchy-liquid-glass-theme ⭐15 | Glass theme for Omarchy — blur, squircles, per-app alpha, palette sampled from its own wallpaper by an OKLCH harmonizer | Shell · Python · Hyprland |
| sih26038 | Explainable DR screening for rural India (SIH 2026). Calibrated ICDR grading, Grad-CAM checked against two independent lesion channels, policy that escalates rather than guessing. Val sens 0.982 / spec 0.917, Wilson CIs reported | MATLAB · Simulink · ResNet-50 |
| Ayur-Ai | Fully offline Ayurvedic leaf classifier + TF-IDF herb retriever. CV acc 0.9864, INT8 PTQ lossless at 4× smaller. No LLM, runs on-device | PyTorch · scikit-learn · Streamlit |
| agent-spend-policy-engine | Policy engine that decides whether an autonomous agent may spend, before payment. Real x402 settlement on Algorand testnet, hash-chained audit ledger anchored on-chain | FastAPI · Algorand · Pydantic |
| aidevops | Vulnerability triage that ranks live Kubernetes workloads by real exposure — EPSS/KEV signals, in-cluster deployment, ports-and-adapters core | Kubernetes · FastAPI · Prometheus |
| API-Pulse | API testing and monitoring dashboard — analytics, threshold alerts, JWT auth, SSRF protection, Docker | Flask · PostgreSQL · Redis |
| ALT-Group-08 | Ranks candidates against one open role and explains every position — advisory only, no auto-reject, with redaction and counterfactual checks | Python · Flask · Pydantic |
Languages · Python · MATLAB · JavaScript · SQL · Shell
ML · PyTorch · scikit-learn · ResNet · Grad-CAM · TF-IDF · post-training quantization · Streamlit
Backend · FastAPI · Flask · SQLAlchemy · Pydantic · Redis · PostgreSQL
Infra · Docker · Kubernetes · Prometheus · pytest · uv
Every project above keeps a written record of what was tried and rejected, not just what
shipped — Ayur-Ai/experiments/journal.md
logs the three optimizations that failed and the ceiling they hit, and sih26038 records
the acceptance criteria it did not meet. I'd rather publish the negative result than a
number nobody can reproduce.
LinkedIn · linkedin.com/in/jitheswar | Location · Hyderabad, India