I build small, verifiable tools for the parts of the AI stack that fail quietly — agent sandboxes, MCP gateways, model-fit predictors, and post-quantum crypto. Each of the flagship repos below ships with a real test suite, CI, and — where it makes a claim about speed or correctness — a committed benchmark you can re-run yourself. It's the latest chapter of a multi-year arc: coursework → CV/security research → hackathons and GenAI → shipping infrastructure.
- 💼 Now: AI/ML Engineering Intern @ RnR Consulting (Delhi) — shipped a model-routing harness that cut inference cost 58% / 65%; Go microservices in a 29-service, Temporal-orchestrated backend serving 500+ concurrent users
- 🎓 Research: 1 Springer Q1 journal · 3 IEEE / SCOPUS papers · 2 book chapters — deepfake detection, federated learning, PQC (see the Research & Impact panel below)
- 🧭 Focus: agent security · AI infrastructure · post-quantum readiness · backend systems
- 🧪 How I work: reference-validated implementations, adversarial tests, reproducible benchmarks — no unverified claims
- 🎓 B.Tech (Hons.) CS, Data Science — Symbiosis Institute of Technology, Nagpur (2022–26), CGPA 8.5/10
- 📫 Reach me: LinkedIn · Portfolio · Email
The board above is regenerated daily from the live GitHub API — CI dots, uptime bars, versions and "last commit" ages are real. When a flagship's build is red, it says so.
| Repo | Stack | Release | What it does |
|---|---|---|---|
| agent-rules-audit | JavaScript | — | Static linter for AI-agent rule files (Cursor / Claude / Copilot) — flags over-broad tool grants and injection-prone instructions. |
| mcp-sentinel | Rust | — | Offline risk scanner for MCP client configs — grades each server A–F on inline secrets, @latest pins, shell indirection, typosquats. |
| toolcage | Rust | v0.1.0 |
WASM sandbox for MCP tool calls — a fresh wasmtime Store per call, deny-by-default caps, HMAC-signed tools/list pagination. |
| agent-flightbox | Go | v0.1.0 |
Flight recorder for agent processes — captures the syscall / exec / network surface of a run to tamper-evident JSONL, with a session diff. |
| Repo | Stack | Release | What it does |
|---|---|---|---|
| mcp-gateway-lite | Go | v0.4.0 |
Single-binary reverse proxy for MCP — allowlist filtering, rate limiting, tamper-evident audit log, tools_lock against rug-pulls. |
| modelgate | Go | — | Multi-provider LLM gateway — routing, automatic fallback, token/cost accounting, metadata-only audit trail. stdlib-only. |
| localmodel-fit | Go | v0.1.0 |
Predicts whether a GGUF model fits and how fast it decodes on given hardware — MoE-aware, validated against real ollama runs. |
| trace2eval | JavaScript | — | Turns raw agent traces into scrubbed, deduplicated eval datasets — PII scrub before dedupe, deterministic, offline. |
| Repo | Stack | Release | What it does |
|---|---|---|---|
| ml-kem-rb | Ruby | — | Reference ML-KEM (FIPS 203) in pure Ruby, plus a real hybrid X25519 + ML-KEM-768 KEM implementing the TLS 1.3 draft wire format. |
pqc-scan 🔒 private |
Rust | — | Crypto inventory → CycloneDX CBOM → A–F post-quantum readiness grade, with live TLS 1.3 handshake checks. Launches Sept 2026. |
| Repo | Stack | Release | What it does |
|---|---|---|---|
| idempotent-rack | Ruby | v0.1.0 |
Idempotency-Key middleware for Rack/Rails — dedupes retried POST/PUT against a pluggable store. (0.3.0 Redis/ActiveRecord backends in progress.) |
| acts-as-mcp | Ruby | — | Expose ActiveRecord models as MCP tools from a Rails app with one class macro — scoped, read-only-by-default agent access. |
| gemfile-lock-audit | Ruby | — | Audits a Gemfile.lock for yanked gems, git-sourced deps, and pins that drift from the lockfile — zero network, CI-friendly. |
The recent flagships are the current chapter — the portfolio goes back to 2023. A few older public repos worth surfacing:
| Repo | Year | What it is |
|---|---|---|
| Real-and-fake-face-distinction | 2024 | Keras CNN classifying real vs AI-generated faces — the research behind the SCOPUS-indexed IEEE deepfake-detection paper (93.5% accuracy). |
| pbl | 2024 | Deep-learning image encryption — DCGAN key generator + attention/residual CNN, with NPCR/UACI security analysis. |
| GigX | 2025 | Decentralized freelance marketplace — Solidity/Hardhat on-chain jobs, escrow & reputation + Next.js. 3rd place, BITS Pilani Web3.0 '25. |
| GenAI-Platform | 2025 | GenAI workspace UI: multi-PDF RAG chat, GraphRAG entity graphs, ATS resume feedback, text-to-SQL (React · Supabase). |
Peer-reviewed research is rare on a GitHub profile — 1 Springer Nature Q1 journal, 3 SCOPUS-indexed IEEE papers (ICISCT '24 @ Kookmin University, ICPCT '25 @ Amity), and 2 book chapters (federated learning; renewable-energy AI). The deepfake detector above fed one of those IEEE papers.
Reproduce these numbers yourself ▸
# mcp-gateway-lite — reverse-proxy overhead (Apple M4, go1.26.5)
go test -run '^$' -bench . -benchtime=2s ./gateway/... # 29.3µs direct vs 69.9µs through gateway
# toolcage — per-call WASM sandbox overhead (ubuntu-latest CI, 200 echo calls)
python3 ci/bench.py WORK ./target/release/toolcage x.wasm 200 # 0.415ms median vs 0.089ms unsandboxed floor
# localmodel-fit — prefill 1/params scaling (Apple M4, real ollama)
go run ./bench -model qwen2.5:0.5b -hw m4 -params 494032768 # measured 0.5b/1.5b ratio 3.06–3.18 vs exact 3.125
# ml-kem-rb — hybrid X25519+ML-KEM-768, TLS draft wire format (FIPS 203)
ruby -rml_kem/hybrid -e 'p MLKem::Hybrid.client_init[0].bytesize' # => 1216 (server share 1120, shared secret 64 B)Two of my repos (ml-kem-rb, pqc-scan) exist because the crypto deadlines below are real and close. The clock counts down live against these US federal dates (NIST / NSA primary sources).
These aren't scattered side-projects.
modelgategates the LLM calls;mcp-gateway-litefilters the tool calls;toolcagesandboxes each one;mcp-sentinel,agent-rules-audit,agent-flightboxandtrace2evalwatch the run — andml-kem-rb/pqc-scanare the post-quantum floor the whole thing has to stand on.
Node size is real commit count. The
MCPhub links the repos that actually speak the protocol; the bright edge is the CI cross-check where mcp-gateway-lite verifies mcp-sentinel's own lockfile output — a real test, not a diagram flourish.
── EOF ────────────────────────────────────────────────────────────
if a tool makes a claim, it ships with the test that proves it.
────────────────────────────────────────────────────────────────────
This whole page is a program. Nine custom SVG instruments, built from live GitHub data by assets/generate.py, committed to this repo, and refreshed every day by a GitHub Action. No third-party widget services. No mocked numbers. Every figure is real and reproducible — down to the red build I haven't hidden.
◆ self-hosted ◆ live-sourced ◆ dark/light aware ◆ animated in-SVG ◆ zero external widgets