Skip to content
View Aditya365x's full-sized avatar

Highlights

  • Pro

Block or report Aditya365x

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. 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
Aditya365x/Readme.md

Hi, I'm Aditya Raj 👋

AI Engineer · Agentic Systems · RAG · LLM Fine-Tuning

B.Tech @ NIT Rourkela (Minor in CSE, 2028) · Lead AI Engineer @ Atlas AI Labs · Secretary, ML4E Club

LinkedIn Email LeetCode


🧭 About me

I build AI systems that run like production services, not notebook demos: agents that plan and use tools, retrieval pipelines that check their own sources, and fine-tuned models that ship behind an API.

  • 🤖 Agents: LangGraph state machines, MCP servers and clients, tool use, self-review loops
  • 🔎 Retrieval: hybrid BM25 + vector search, rank fusion, reranking, corrective RAG, retrieval evals
  • 🧪 Models: QLoRA fine-tuning, CNNs, gradient-boosted models, evaluation and guardrails
  • 🚀 Shipping: FastAPI, Docker, CI/CD, cloud deploys on Azure, AWS, Vercel and Hugging Face
  • 🎯 Currently looking for AI Engineering / SDE internships for Summer 2027

⭐ Featured projects

An agent that plans a task, searches the web, reads pages and PDFs, runs Python in a locked-down sandbox, searches your documents, reviews its own answer and streams every step live. Also an MCP server and client.

  • LangGraph plan → act → reflect loop, multi-provider routing with fallbacks
  • Hybrid pgvector + BM25 retrieval with RRF, measured by recall@k / MRR / nDCG
  • Guardrails, budgets, Prometheus metrics, 53 tests + 2 eval suites in CI
  • Live demo on Azure Container Apps

LangGraph FastAPI LiteLLM pgvector Redis MCP Next.js Docker

Upload a PDF and ask questions. CRAG grades every retrieved passage and falls back to web search when the document can't answer, with [p.N] / [web N] citations on every sentence.

  • Hybrid BM25 + MiniLM retrieval fused with reciprocal rank fusion
  • LLM relevance grading → CORRECT / AMBIGUOUS / INCORRECT routing
  • Sentence-level filtering and a transparent pipeline trace per answer
  • Live demo on Vercel

LangGraph FAISS BM25 FastAPI Tavily Llama 3.3 70B

Full-stack farm assistant: yield prediction, leaf-disease detection with RAG-based treatment advice, a multilingual agronomy chatbot grounded in oilseed research, and live IoT sensor dashboards.

  • CatBoost yield model, Keras fertilizer recommender, CNN disease classifier
  • "Krishi Saathi" history-aware RAG chatbot over a library of oilseed research PDFs (FAISS)
  • FastAPI + JWT backend, Next.js web app, Flutter mobile app

FastAPI CatBoost TensorFlow LangChain FAISS Next.js Flutter

Turns plain-English questions into SQL with a fine-tuned 1B-parameter Gemma model.

  • QLoRA (4-bit NF4, LoRA rank 8) on gemma-3-1b-it, ~15 MB adapter
  • Trained on the schema-aware sql-create-context corpus
  • Two-tier deploy: FastAPI backend → Hugging Face inference Space

PyTorch QLoRA PEFT Hugging Face FastAPI Docker

The allocation layer between a disaster alert and the teams on the ground: incident clustering, shelter and supply allocation, routing and live dashboards for district control rooms.

FastAPI PostgreSQL WebSockets Docker Optimization

Three specialized agents (Researcher → Analyst → Writer) collaborate through shared LangGraph state to produce structured research reports.

LangGraph LangChain FastAPI Docker


🛠️ Tech stack

Area Tools
Languages Python · C++ · SQL · TypeScript
GenAI & Agents LangChain · LangGraph · MCP · LiteLLM · Hugging Face · OpenAI · CrewAI
ML / DL PyTorch · TensorFlow/Keras · scikit-learn · CatBoost · QLoRA/PEFT · MLflow
Retrieval & Data PostgreSQL + pgvector · ChromaDB · FAISS · BM25 · Redis
Backend & DevOps FastAPI · Docker · GitHub Actions · Azure Container Apps · AWS · Vercel
Frontend Next.js · React · Tailwind CSS

🏆 Highlights

  • 🥉 Smart India Hackathon 2025 — National 2nd Runner-Up: crop-yield prediction model (R² 0.93) on 70K+ samples and a CNN leaf-disease classifier
  • 💻 400+ LeetCode problems solved, focused on dynamic programming and graphs
  • 📜 IBM Agentic AI & Workflows Specialization (Coursera): LangGraph, CrewAI, AutoGen
  • 👥 Secretary, ML4E (Machine Learning for Everyone), NIT Rourkela: run hands-on LLM and agent workshops for a 60+ member community and co-organize inter-club hackathons

📫 Open to AI Engineering / SDE internships and collaborations. Reach me at adityaraj.adi365@gmail.com

GitHub stats

Pinned Loading

  1. AgriSutra AgriSutra Public

    AI farm assistant for oilseed farmers: CatBoost yield prediction, CNN leaf-disease detection with RAG advice, agronomy chatbot and IoT monitoring. FastAPI, Next.js, Flutter.

    Jupyter Notebook

  2. CRAG CRAG Public

    Corrective RAG: ask any PDF. Hybrid BM25 + vector retrieval, LLM relevance grading and web-search fallback, with page-level citations. LangGraph + FastAPI.

    Python

  3. multi-agent-collaboration-system multi-agent-collaboration-system Public

    Multi-agent research pipeline: Researcher, Analyst and Writer agents collaborate through shared LangGraph state to produce structured reports.

    HTML

  4. stockAnalyser stockAnalyser Public

    Python

  5. Text-to-SQL-Generator_LLM_Fine_tuned Text-to-SQL-Generator_LLM_Fine_tuned Public

    Natural language to SQL with Gemma-3-1B fine-tuned via QLoRA (4-bit NF4, LoRA r=8). FastAPI backend + Hugging Face Space deployment.

    Jupyter Notebook

  6. atlas-agent atlas-agent Public

    Autonomous AI agent: plans tasks, browses the web, runs sandboxed Python and searches your docs with hybrid RAG. MCP server + client. FastAPI, LangGraph, pgvector, Next.js.

    Python