Currently focused on Multi-Agent Systems, AI Infrastructure and Production ML
Sentinel — Multi Agent AI code auditor
AI engineer that audits codebases with multi-agent LangGraph orchestration and opens validated, human-reviewed fix PRs — never auto-merges
Repository: https://github.com/AlchemistReturns/Sentinel
- Built a multi-agent LangGraph system (Security, Quality, Test analysts) that audits codebases in parallel, grounded in real tool output (semgrep, pip-audit) rather than LLM pattern-matching.
- Implemented a diagnose-propose-validate-PR pipeline: findings get deterministic risk tiers, fixes are validated via lint/test regression checks before opening GitHub PRs, with risky changes forced to draft for human review.
- Hardened for production with a Redis-backed task queue, distributed locking to prevent concurrent duplicate fixes, per-audit spend caps, and a kill switch — all verified under real concurrent load.
+------------------------+
| Next.js Dashboard |
+-----------+------------+
|
HTTPS / WebSocket
|
+-----------v------------+
| FastAPI Orchestrator |
| (REST + WebSocket) |
+-----------+------------+
|
+-----------v------------+
| LangGraph Audit Graph |
+-----------+------------+
|
+-----------v------------+
| Ingest / Scope |
+-----------+------------+
|
+-------------------------+-------------------------+
| | |
+---------v--------+ +----------v---------+ +---------v---------+
| Security Analyst | | Quality Analyst | | Test Analyst |
| (semgrep,pip-audit)| | (dead code) | | (coverage gaps) |
+---------+--------+ +----------+---------+ +---------+---------+
| | |
+-------------------------+-------------------------+
|
+-----------v------------+
| Diagnose (risk tier, |
| dedup via semantic cache)|
+-----------+------------+
|
+-----------v------------+
| Propose Fix (LLM) |
+-----------+------------+
|
+-----------v------------+
| Validate (ruff+pytest) |
+-----------+------------+
|
+-----------v------------+
| Open PR (GitHub API) |
| draft if risky |
+------------------------+
+---------------------+ +----------------------+
| Postgres + pgvector |<----->| LangGraph Checkpointer|
+---------------------+ +----------------------+
+---------------------+
| Redis: queue, cache,|
| lock, kill switch, |
| spend budgets |
+---------------------+
+---------------------+ +----------------------+
| OpenAI (LLM calls) | | LangSmith (tracing) |
+---------------------+ +----------------------+
HealthX — AI-powered telehealth platform
AI-powered telehealth platform built for underserved rural communities with unreliable connectivity.
Repository: https://github.com/AlchemistReturns/HealthX
- Built HealthX, a full-stack telehealth platform (React/Vite, Node.js/Express, Firebase Auth) for underserved rural regions with unreliable connectivity, integrating Google Gemini for context-aware symptom analysis using patient history.
- Designed a hybrid online/offline architecture with client-side image compression and a hardcoded fallback dataset, ensuring core diagnosis and hospital-lookup features remain functional on unstable 2G/3G networks.
- Implemented a GPS-based hospital finder combining live OpenStreetMap/Overpass queries with an offline emergency database, plus one-tap Google Maps/Waze navigation handoff.
+----------------------+
| React App |
+----------+-----------+
|
+---------------------+----------------------+
| |
| Internet Available | Offline
| |
+-------v--------+ +---------v---------+
| Express API | | Local Emergency |
+-------+--------+ | Dataset |
| +---------+---------+
+-------+---------+ |
| Firebase Auth | |
+-----------------+ |
| |
+-------v---------+ +--------v--------+
| Gemini API | | Offline Search |
+-----------------+ +-----------------+
|
+-------v------------------+
| OSM / Overpass API |
+--------------------------+
|
Google Maps / Waze
Valerix — Fault-tolerant microservices platform
Fault-tolerant distributed e-commerce backend with graceful degradation.
Repository: https://github.com/rawadhossain/Valerix
- Designed a fault-tolerant Node.js/Express microservices architecture (Order, Inventory, API Gateway) implementing timeout-triggered fallback from synchronous HTTP calls to asynchronous RabbitMQ messaging, preventing request hangs during downstream service degradation.
- Engineered idempotent message processing via an
IdempotencyLogtable to guarantee exactly-once inventory deduction under message retries/duplication, alongside indefinite connection-retry logic for self-healing broker reconnection on startup or outage. - Instrumented services with Prometheus (RED metrics) and Grafana dashboards, and built real-time latency-alerting in the frontend, enabling observability into distributed failure modes (simulated via a configurable latency-injection mechanism).
+-------------------+
| Frontend |
+---------+---------+
|
API Gateway
|
+------------+------------+
| |
HTTP Success Timeout
| |
+--------v--------+ +-------v--------+
| Order Service |------->| RabbitMQ |
+--------+--------+ +-------+--------+
| |
| Retry / Queue
| |
+--------v--------+ +-------v--------+
| Inventory |<-------| Worker |
| Service | +-------+--------+
+--------+--------+ |
| |
+--------v---------+ |
| Idempotency Log | |
+------------------+ |
|
+--------------------------+
|
Prometheus ---> Grafana ---> Alerts
Grit — Local-first Git intelligence CLI
Local-first Git intelligence CLI for engineering decision tracking.
Repository: https://github.com/AlchemistReturns/grit
- Built grít, a Go CLI that hooks into git pre-commit, post-rewrite, and post-commit stages to run adaptive interviews, capturing the reasoning behind commits, dependency changes, and reverts.
- Implemented a real-time file watcher with a keyword-based complexity scorer, vague-identifier detection, and paste/AI-assist heuristics, all configurable via
.grit.yamlthresholds. - Designed a local-first SQLite (WAL mode) storage layer with per-repo isolation, plus
stats,decision(ADR-style), and Markdown/JSONpushexport commands — no cloud dependency.
Developer
|
Git Commit
|
+-------------+--------------+
| |
pre-commit post-commit
| |
+-------------+--------------+
|
Grit Engine
|
+--------------+----------------------+
| | |
File Watcher Interview Engine Complexity Analysis
| | |
+--------------+----------------------+
|
SQLite (WAL Mode)
|
+-------------+-------------+
| | |
stats decision push
|
Markdown / JSON Export
CurioKids — AI-assisted educational platform
AI-assisted educational platform for early learners.
Repository: https://github.com/AlchemistReturns/CurioKids
- Built CurioKids, a cross-platform React Native/Expo educational app with dual parent/child dashboards, an interactive lesson engine, and gamified mini-games (math, literacy, motor skills) for early learners.
- Designed a Node.js/Express backend with Firebase Authentication and Firestore, syncing progress in real time to give parents actionable analytics and drive point-based rewards and leaderboards.
- Integrated an AI chatbot via a self-hosted RAG service (FastAPI + Ollama, Dockerized), enabling local LLM-backed learning assistance without external API dependencies.
React Native App
|
+------------------+-------------------+
| |
Parent Dashboard Child Dashboard
| |
+------------------+-------------------+
|
Node.js / Express
|
+-----------------+----------------+
| |
Firebase Auth Firestore
| |
+-----------------+----------------+
|
FastAPI RAG Service
|
Ollama LLM
"Intelligence is the system. I build what holds it up."
