Skip to content
View Ezed9's full-sized avatar

Highlights

  • Pro

Block or report Ezed9

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
Ezed9/README.md

divider

⚡ whoami // Systems & AI Engineering

engineer:
  name: "Nishit Baishya"
  handle: "@Ezed9"
  education: "B.Tech in Electronics & Communication Engineering @ NIT Silchar (2023–2027)"
  location: "Assam, India"

current_focus:
  - "Co-Founder & Full-Stack Lead @ Maxweb (max-web.studio) — Production Healthcare & B2B Systems"
  - "Deterministic AI Agent Security & Model Context Protocol (MCP) Guardrails"
  - "High-Throughput Distributed Queues, Transactional Outboxes & Rate-Limited Worker Pools"
  - "Generalizable Multimodal Representation Learning (CLIP ViT-L/14 Synthetic Media Detection)"

divider

🏗️ Live System Architecture // Flagship Projects

System Architecture — Webhook Delivery Platform & Bouncer MCP Proxy

  • Destination Taint Tracking: Built and published a stdio MCP proxy on PyPI that denies an agent's send/share calls to destinations it first read in untrusted tool output. Plain-Python rules, no LLM in the decision path.
  • Deterministic Contract Engine: Enforces schema pinning, per-session call budgets, and path-traversal constraints before tool execution.
  • Benchmarked with its cost: On AgentDojo (gpt-4o-mini, 164 attack cases) the attacker's goal was reached in 48 cases without Bouncer and 0 of 163 with it, but Bouncer allowed no send at all, so email/share tasks went 4/6 → 0/6. One fixed, non-adaptive attack. 150 tests in GitHub Actions CI, PyPI Trusted Publishing.

  • At-Least-Once Semantics: Engineered idempotent event ingestion (Idempotency-Key unique deduplication) paired with transactional-outbox enqueueing.
  • Concurrent Worker Pool: Implemented using PostgreSQL SELECT FOR UPDATE SKIP LOCKED, HMAC-SHA256 payload signing, exponential-backoff-with-jitter retries, and dead-letter queueing (DLQ).
  • Noisy-Neighbor Isolation & Scale: Built per-tenant Redis token-bucket rate limiting and structured telemetry; sustained 315 events/s throughput, 100% delivery success, and 5 ms p95 latency under 30% receiver failure stress across 50 concurrent Locust clients.

  • Cross-Generator Evaluation: Benchmarked synthetic-face detectors across 5 generative architectures (StyleGAN, Stable Diffusion, SDXL, Flux, DALL-E 3) under Dr. Debbrota Paul Chowdhury at NIT Silchar.
  • Generalization Findings: Demonstrated that pretrained CLIP ViT-L/14 representations generalize significantly better to unseen generators than CNN baselines (0.92 vs. 0.84 AUROC).
  • Vulnerability Discovery: Exposed how modern text-to-image synthesis pipelines cause conventional detector accuracy to collapse below chance (AUROC < 0.50).

  • Critical Healthcare Deployment: Architected and deployed a production OPD/IPD hospital patient registration platform sized for 500+ patients/day across 5 concurrent counters.
  • Operator-First Ergonomics & Privacy: Built keyboard-only rapid data entry, strict server-side Pydantic validation, instant printable registration slips, and zero-PII audit logging.
  • B2B & AI Automation: Delivered SSR production web apps across 3+ B2B clients at max-web.studio and engineered a Python + LLM API sales-automation agent.

divider

💼 Experience

Role Organization Period Key Impact
Co-Founder & Full-Stack Lead Maxweb (max-web.studio) Jan 2026 – Present Architected a 500+ patients/day hospital OPD/IPD registration system across 5 concurrent counters; delivered SSR web platforms for 3+ B2B clients and built an LLM-powered sales-automation agent.
Research Intern — AI Face Detection NIT Silchar (Advisor: Dr. D. P. Chowdhury) Summer 2026 Evaluated deepfake detectors across 5 generators (StyleGAN to DALL-E 3), proving CLIP ViT-L/14 feature superiority (0.92 AUROC) over CNN baselines (0.84 AUROC).
Industrial Trainee (IS & Telecom) Indian Oil Corporation Ltd. (IOCL), Digboi Refinery Jun – Jul 2026 Analyzed critical-infrastructure enterprise networking: dual-carrier MPLS-VPN WAN (Cisco Nexus/Catalyst), SolarWinds telemetry (~99.99% WAN availability), and ISO 27001 IT/OT network segregation.

divider

🛠️ Tech Stack & Systems Arsenal

Tech Stack Icons

Domain Technologies & Core Competencies
Languages Python • C++ • TypeScript • JavaScript • SQL
Backend & Distributed Systems FastAPI • REST APIs • AsyncIO • Transactional Outbox • Idempotency • Token-Bucket Rate Limiting • System Design
Databases & Storage PostgreSQL (SKIP LOCKED, Indexing, Query Optimization) • Redis • SQLAlchemy • Alembic
AI/ML & Agent Security Model Context Protocol (MCP) • Prompt-Injection Defense • PyTorch • LangChain • LangGraph • RAG • Gemini & LLM APIs
Frontend & Web Next.js (SSR) • React • TypeScript • Tailwind CSS • HTML5 / CSS3 • Responsive UI
DevOps, CI/CD & Testing Docker • Docker Compose • GitHub Actions CI/CD • Pytest • Locust Load Testing • Linux • Git

divider

📈 GitHub Telemetry & Contribution Snake

Ezed9 GitHub Stats Ezed9 GitHub Streak

github contribution grid snake animation

divider

Built with precision by Nishit Baishya (@Ezed9) • max-web.studio • nishitbaishya9@gmail.com

Pinned Loading

  1. mcp-bouncer mcp-bouncer Public

    A deterministic exfiltration firewall for MCP tool calls — gates the destination, not the tool. No LLM in the decision path.

    Python 2

  2. webhook-delivery webhook-delivery Public

    HTML

  3. faceProof faceProof Public

    Benchmark of AI-generated face detectors: CLIP ViT-L/14 vs CNN baselines across 5 generators — robustness, calibration (AUROC/ECE), with a public Gradio demo

    Jupyter Notebook