I'm a B.Tech CSE (AI & ML) student at Sikkim Manipal Institute of Technology (SMIT), building across applied AI/ML, software engineering, and cloud-native systems — and learning in the open. I like turning fuzzy problems into tools that detect, decide, and deploy, and I care about evidence, correctness, and clean engineering over hype.
_%C2%B7_SMIT-7C3AED?style=flat-square){kind=icon}
|
Computer vision, model training, and practical inference pipelines that hold up outside a notebook. |
Evidence-grounded answers with citations — and honest abstention when the evidence is thin. |
|
Tamper-evident, offline-first integrity checks and forensic heuristics. |
Containers, CI/CD, and GitOps — shipping software the reproducible way. |
🛰️ SAKSHYAVendor-agnostic CCTV forensic evidence platform — evidence ingestion, AI triage, cross-camera re-ID, deleted-clip recovery, and a tamper-evident cryptographic chain-of-custody. FastAPI + React, CI-green with a full test suite.
|
🔎 RAGBenchAdaptive RAG platform + evaluation harness — grounded answers with citations, abstention when evidence is thin, and retrieval-strategy comparisons (result on the evaluated corpus is neutral/tied, by design).
|
⚙️ OpsForgeCloud-native DevSecOps & SRE platform — code → test → sign → GitOps deploy → observe, with SLOs, burn-rate alerting, tracing, and chaos. Operator-verified on real AWS + a kind cluster.
|
🛡️ Log-GuardianOffline-first log-integrity monitor with forensic heuristics — streaming scans, timestamp-anomaly detection, baseline analysis, tamper-evident commitments, and a CLI.
|
|
Aerial waste intelligence — a fine-tuned YOLOv8 detector that spots illegal dumping from drone footage and scores severity. 🏆 ELCIA Next-Gen Tech Hackathon — 2nd Runner-Up (top 3 of 180+ teams).
|
Production-style Kubernetes GitOps on real AWS — EKS provisioned with Terraform, CI/CD via GitHub Actions + OIDC, Argo CD delivery, Prometheus/Grafana/Loki observability, and demonstrated rollback.
|
Languages
AI / ML · Web
Cloud · DevOps · Tools
I use these at a student-builder level — comfortable working with the core set (Python, FastAPI, React/TS, Docker), and actively exploring cloud-native tooling (Kubernetes, GitOps, MLOps). Also: scikit-learn · NumPy · pandas · Streamlit.
A real, typeable terminal — run help, projects, oss, grade… (GitHub READMEs can't run JS, so it lives on GitHub Pages.)
Merged into upstream projects 🟢
| Project | What I fixed | PR |
|---|---|---|
| mlflow/mlflow | format_prompt was replacing placeholders that appear inside variable values |
#26807 |
| ultralytics/ultralytics | Skip torch.compile on CUDA when Triton is unavailable |
#26701 |
| kubernetes-client/python | Relax the certifi lower bound and stop Dependabot raising it |
#2745 |
| cloud-custodian/cloud-custodian | Skip ROLLBACK_COMPLETE stacks in AWS CFN tag actions |
#11215 |
| aws-cloudformation/cfn-lint | Validate --registry-schemas types against their schema |
#4763 |
| aws-cloudformation/cfn-lint | Check the length of a Description that uses a dynamic reference |
#4765 |
In review (open PRs) 🟡
- KServe — Python-client fixes: #6405 · #6407 · #6414 · #6521
- mlflow/mlflow — always load model config as UTF-8: #27008
- kubeflow/docs-agent — #283 · #285
All links are public PRs — browse everything on my profile.
Activity grade from a transparent rubric (commits · PRs merged · collaboration breadth · sustained contributions) — see the live calc via grade in the interactive terminal. Not an academic grade.
