I'm an AI/ML Engineer and Backend Systems Developer who builds things from scratch — not just to ship, but to understand how they work internally.

I didn't just use LLMs — I built a RAG pipeline with LangChain, chaining retrieval, embeddings, and LLM calls into a working Q&A system over custom data. I didn't just use Redis — I rebuilt it in C++, stress-tested it with 10,000 threads, and understood why single-threaded execution is a design choice, not a limitation. That's how I think.

• 🤖 AI/ML Focus: LangChain · RAG Pipelines · Custom CNNs · Medical Imaging · LLMs
• ⚙️ Systems Focus: Concurrency · Caching · Fault Tolerance · Performance
• 🔬 Research: Published paper on ML model behavior (IJSET, Vol. 13, 2025)
• 🏗️ Approach: Understand the internals before touching the scale knob
• 🎯 Goal: AI/ML Engineer or SDE-1 role — building real, production-grade systems

"Built a full retrieval-augmented generation system — because using ChatGPT is easy, but understanding how to wire your own data into an LLM pipeline is where the real learning is."

What I built:

• Document ingestion pipeline → chunking → embedding model → vector store
• LLM-powered Q&A over custom knowledge bases with context injection
• Prompt engineering for structured, reliable outputs from LLM calls
• Chained LangChain components: loaders → splitters → retrievers → LLM → output parsers

Python · LangChain · RAG · Embeddings · Prompt Engineering · LLMs

"Reduced CT scan manual review from 15–20 minutes to under 5 seconds. This is what applied AI looks like."

What I built:

• Complete ML pipeline: data preprocessing with OpenCV + NumPy, class imbalance handling, augmentation
• Custom CNN architecture: design → training → evaluation → deployment
• Binary classification: malignant vs. benign CT scan images
• Live demo →

Python · TensorFlow · Keras · OpenCV · NumPy · Medical Imaging · CNN

What I built:

• End-to-end CV app: face detection → background removal → government-spec compliance validation
• Custom CNN classifier checks size, lighting, and face positioning against official standards
• Cuts passport photo prep from 30+ min (studio visit) to under 30 seconds
• Real users, real product — deployed and live

Python · OpenCV · CNN · React · Computer Vision · Vercel

"Built this to understand why Redis chose single-threaded execution. Turns out it's a deliberate design choice for predictability — not a limitation."

Architecture:

Client Threads (N)                Worker Thread (1)
──────────────────                ─────────────────
thread_1 ──┐
thread_2 ──┼──► Command Queue ──► Serial Executor ──► Response
thread_N ──┘    (mutex-locked)    (no data races)     (future/promise)

What I built:

• Single-worker command queue — all mutations serialized, zero coarse locking
• SET / GET / DEL across N concurrent clients via std::mutex + std::condition_variable
• TTL-based expiration with std::chrono::steady_clock + LRU eviction
• Stress tested: 10,000 concurrent threads — zero data corruption

C++ · Concurrency · Thread Safety · LRU · TTL · System Design

Deployed in production at CT University — running across 200+ network-connected endpoints.

What I built:

• Long-running OS-managed background service with persistence across reboots
• Priority-aware, conflict-free job scheduling with ±1s accuracy across 100+ concurrent jobs
• Structured rotating log pipeline + auto-restart failure recovery + Singleton pattern

Python · APScheduler · Multithreading · Fault Tolerance · Structured Logging

• 15+ REST API endpoints — JWT auth, social graph, personalised feed
• Compound-indexed MongoDB, rate limiting, pagination, Cloudinary CDN
• Full production deployment — live at connx.vercel.app

Node.js · MongoDB · JWT · Next.js · REST APIs

• 300+ problems solved across Easy, Medium, Hard — top 17% globally
• Focus areas: Trees, Graphs, DP, Sliding Window, Concurrency problems

Backend / Systems Intern — CTech Labs, Ludhiana (Dec 2023 – May 2024)

• Sole feature owner from requirement → production deployment in Agile environment
• Deployed Python scheduling service across 200+ endpoints — reduced manual overhead by ~100%
• Engineered failure recovery, rotating logs, auto-restart; ±1s scheduling accuracy across 100+ jobs

Full Stack Developer Intern — Suven Consultants (Remote) (Jun 2023 – Aug 2023)

• Built React.js + Node.js client applications; improved page load speed by 20%
• Engineered modular REST APIs integrated with MongoDB in Agile team environment

"I prefer understanding how systems work internally — before scaling them externally."

• Build to understand — I've rebuilt Redis in C++ and built my own RAG pipeline from scratch. Using a tool you built teaches you what no docs can.
• AI that actually works — not notebooks, but deployed products with real users (PassportPro, Lung Cancer CNN)
• Correctness before performance — a fast broken system is worse than a slow correct one

Actively looking for AI/ML Engineer or SDE-1 / Backend Engineer roles. If you're building in AI, infrastructure, fintech, or distributed systems — let's talk.

AI that works. Systems that scale.