I am a Software Development Engineer (SDE) at DailyKaam and a Computer Science & Engineering student at NIT Trichy. I focus on backend API design, distributed database architectures, cloud deployments, and integrating Generative AI pipelines. My engineering philosophy revolves around writing clean, testable code, optimizing execution performance, and building systems that scale efficiently under production loads.

• Production Web Application Development: Architect client-facing modules using React and Next.js, reducing rendering latencies and improving load times.
• Mobile Application Development: Build and maintain cross-platform mobile app configurations to ensure UI parity and performance.
• Backend API Engineering: Design, implement, and optimize RESTful backend API routes in Node.js and Express to manage high-volume data streams.
• Database Design & Optimization: Model relational and non-relational database schemas in PostgreSQL and MongoDB, implementing indexing and normalized layouts to reduce lookup latency.
• Deployment & Production Support: Maintain Docker runtimes, configure continuous integration scripts, and support live production rollouts.
• System Scalability & Performance Optimization: Audit and resolve bottleneck routes to ensure low network latency and maximum request throughput.

• High-Throughput Scale: Architected a MERN-based M-Commerce platform (Live Application) with Progressive Web App (PWA) support, scaling container workloads on AWS to support 10,000+ concurrent users.
• Caching & Database Tuning: Implemented Redis key-value caching, relational index strategies, and collection sharding, resulting in a 35% query performance improvement and a 40% reduction in server CPU load.
• Automation Pipelines: Structured automated unit testing and build verification pipelines using GitHub Actions, cutting staging release cycles by 40%.

• Workflow Acceleration: Configured LangChain agentic state-machines and multi-model routing loops, boosting administrative operational efficiency by 25%.
• RAG System Engineering: Designed and implemented Retrieval-Augmented Generation (RAG) pipelines (Live Platform) using vector embeddings, Pinecone indices, and semantic search, raising output accuracy by 18%.
• Development Velocity: Created prompt orchestration helpers that shortened overall developer iteration cycles by 30%.

• Business Impact: Automates candidate resume analysis, yielding parsed scores and line-by-line feedback in under 3 seconds.
• Technical Challenges: Mitigating prompt instruction drifts and handling large-context resume documents under rate limits.
• Architecture: Next.js client routes requests to serverless API handlers. Sessions are tracked via Prisma ORM to Neon PostgreSQL, while resume uploads are audited via Gemini API.
• Tech Stack: Next.js, TypeScript, Gemini API, Prisma ORM, Neon PostgreSQL, TailwindCSS.
• Repository Link: DEEPAK-317/DDRAI-CAREER-COACH
• Architecture Flow:

graph TD
    User([User / Candidate]) -->|Uploads Resume| NextJS[Next.js Client]
    NextJS -->|Proxies Request| NextAPI[Next.js Backend API]
    NextAPI -->|Query Auth & Session| Prisma[Prisma ORM]
    Prisma -->|Read/Write State| DB[(Neon PostgreSQL)]
    NextAPI -->|Prompt & Resume Stream| Gemini[Gemini 1.5 Pro LLM]
    Gemini -->|Line-by-Line Critique| NextAPI
    NextAPI -->|Structured JSON Response| NextJS
    NextJS -->|Dynamic Interactive UI| User

• Business Impact: Powers context-aware query processing over massive PDF log archives, returning precise reference citations.
• Technical Challenges: Managing overlap chunk boundaries to maintain semantic intent during document partitioning.
• Architecture: FastAPI server coordinates document parsing, Mistral vector embed indexing, and similarity searches inside Pinecone using MMR retrieval patterns.
• Tech Stack: Python, FastAPI, LangChain, Pinecone Vector DB, Mistral API.
• Links: Repository | Live Demo
• Architecture Flow:

graph TD
    Docs[[High-Volume PDF Logs]] -->|Batch Load| FastAPI[FastAPI Server]
    FastAPI -->|Semantic Text Partitioning| Chunks[Document Chunks]
    Chunks -->|Mistral Embedding API| Embeddings[Dense Vectors]
    Embeddings -->|Store & Index| Pinecone[(Pinecone Vector DB)]
    UserQuery([User Query]) -->|FastAPI semantic lookup| Pinecone
    Pinecone -->|Diverse Retrieval via MMR| RefDocs[Top-K Reference Vectors]
    RefDocs -->|Context Synthesis Prompt| MistralLLM[Mistral AI LLM]
    MistralLLM -->|Coherent Answer with Citations| FastAPI
    FastAPI -->|JSON Stream Response| UserQuery

• Business Impact: Increases audit transparency by generating local explanation weights for transactions flagged as fraudulent.
• Technical Challenges: Balancing high anomaly recall targets while preventing false-positive classification inflation.
• Architecture: Modular machine learning pipeline (Isolation Forest & Random Forest) feeding prediction scores and explanation weights (SHAP) directly to a Streamlit browser dashboard.
• Tech Stack: Python, Streamlit, Scikit-Learn, SHAP, Pandas.
• Repository Link: DEEPAK-317/explainable-fraud-detection

• Business Impact: Automates dynamic email writing with customizable tone categories, speeding up corporate communication flows.
• Technical Challenges: Parsing and validating user inputs into clean prompt formatting constraints.
• Architecture: Streamlit frontend routing payloads to python handler methods interfacing with Google Gemini completions.
• Tech Stack: Python, Streamlit, Gemini API, Prompt Engineering.
• Repository Link: DEEPAK-317/MailMate

• Business Impact: Streamlines scheduling, daily logging, and assignment management for student workflows.
• Technical Challenges: Ensuring session persistence and role validation checks.
• Architecture: Single-page client communicating with Node.js controllers, querying data via Mongoose to MongoDB.
• Tech Stack: JavaScript, Node.js, Express, MongoDB.
• Repository Link: DEEPAK-317/Nitt-Student-Track

• Business Impact: Simplifies ledger tracking and group balances calculation for shared group expenditures.
• Technical Challenges: Designing logic models to process offline transactions and state synchronization.
• Architecture: Cross-platform React Native views pushing updates to Node.js backend endpoints feeding a MongoDB database.
• Tech Stack: TypeScript, React Native, Node.js, Express, MongoDB.
• Repository Link: DEEPAK-317/Split-App (Client) | DEEPAK-317/SplitApp (API)

I contribute to developer utilities and runtime wrappers in the Generative AI and web framework spaces:

• LangChain Ecosystem: Refactored text loading utilities to minimize runtime buffering delays.
• Streamlit Contribs: Optimized dashboard rendering modules for consistent responsive displays.
• rag-cache-optimizer: [Active Project] Building a Python utility for semantic caching of prompt embeddings to reduce LLM latency.

• LeetCode Profile: @deepak_3621
• Statistics: 400+ algorithmic problems solved across Arrays, Dynamic Programming, Graphs, and Trees.
• Competitions: Active participant in weekly optimization rounds to test runtime complexities.

• AWS Certified Cloud Practitioner - Amazon Web Services
• Google Cloud Associate Cloud Engineer - Google Cloud
• Deep Learning Specialization - DeepLearning.AI
• Generative AI with Large Language Models - DeepLearning.AI / AWS

  

• LinkedIn: linkedin.com/in/deepak-kumar-62a4b8270
• Portfolio: deepakkumarnittrichy2026.netlify.app
• Email: deepaknittrichy@gmail.com