IEEE 2026 MTech Projects Using Python — Generative AI & Agentic AI for MTech CS, AI, Data Science & IT Students in Bangalore
Python is the language of modern AI — and in 2025–2026, Generative AI and Agentic AI are the two hottest domains for MTech final year projects. At ProjectsatBangalore, we offer 60+ IEEE 2026 MTech projects using Python across ten cutting-edge AI domains: LLM-based applications with LangChain/LlamaIndex, multi-agent systems with AutoGen/CrewAI/LangGraph, RAG pipelines with vector databases (ChromaDB, FAISS, Pinecone), LLM fine-tuning with LoRA/PEFT/QLoRA, text-to-image and video generation with Stable Diffusion and ControlNet, multimodal AI projects with GPT-4V and LLaVA, AI code generation agents, NLP text generation, AI-powered digital twins and federated LLM with differential privacy. Every MTech Python project includes an IEEE Xplore 2025–2026 base paper, complete Python source code, FastAPI/Streamlit deployment, Docker setup, university-format report (VTU / Anna University / JNTU), architecture diagram, PPT slides and viva Q&A support.
MTech Python Project Domains
- Generative AI — LLM apps, prompt engineering, GPT-4o, Gemini, Claude API
- Agentic AI — AutoGen, CrewAI, LangGraph, ReAct agents, tool-use, reflection
- LLM Fine-Tuning — LoRA, QLoRA, PEFT, instruction tuning, RLHF, DPO
- RAG Pipelines — ChromaDB, FAISS, Pinecone, hybrid retrieval, HyDE, reranking
- Multimodal AI — GPT-4V, LLaVA, CLIP, image+text fusion, video understanding
- NLP & Text Generation — summarisation, translation, sentiment, information extraction
- AI Code Generation — GitHub Copilot-style agents, self-debugging, test generation
- Diffusion Models — Stable Diffusion XL, ControlNet, InstructPix2Pix, LoRA training
- AI Digital Twins — LLM-driven simulation, synthetic data, physics-informed AI
- Federated & Privacy AI — PySyft, differential privacy, secure aggregation
Frameworks & Stack
Python Frameworks Used in Generative AI & Agentic AI MTech Projects
LangChain / LlamaIndex
LangGraph
AutoGen / Microsoft
CrewAI
OpenAI GPT-4o API
HuggingFace Transformers
PEFT / LoRA / QLoRA
PyTorch
Stable Diffusion / SDXL
ChromaDB / FAISS / Pinecone
FastAPI / Flask
Streamlit / Gradio
Docker / Kubernetes
AWS / GCP / Azure
60+ IEEE 2026 Topics
MTech Projects Using Python — Generative AI & Agentic AI Topics
All topics sourced from IEEE Xplore 2025–2026 — IEEE Trans. Neural Networks, IEEE Access, IEEE Trans. Knowledge and Data Engineering, ACM Computing Surveys and Nature Machine Intelligence. Call 9591912372 for topic shortlisting.
Generative AI — LLM Applications
LangChain · GPT-4o · Gemini · Claude · Prompt Engineering · Knowledge Graphs · Synthetic Data
| # | MTech Python Project Title | Tools | Level & Journal |
|---|---|---|---|
| 01 | LLM-Powered Enterprise Knowledge Assistant Using LangChain, GPT-4o and ChromaDB for Multi-Document Q&A with Citation — builds a production RAG chatbot over 1,000+ enterprise PDFs; hybrid retrieval (BM25 + semantic); GPT-4o generates grounded answers with source citations; evaluated on RAGAS faithfulness and relevance metrics. | Python, LangChain, GPT-4o, ChromaDB, FastAPI, React | MTech IEEE Access 2026 |
| 02 | Prompt Engineering Benchmark: Systematic Evaluation of Zero-Shot, Few-Shot, Chain-of-Thought and Tree-of-Thought Prompting on Medical Reasoning Tasks — evaluates 6 prompting strategies across GPT-4o, Gemini 1.5 Pro and Llama 3.1 70B on MedQA and USMLE datasets; ablation study on prompt templates. | Python, OpenAI API, Gemini API, HuggingFace, Streamlit | MTechPhD IEEE Trans. AI 2026 |
| 03 | Synthetic Clinical Trial Data Generation Using GPT-4o and Conditional Variational Autoencoder for Rare Disease Research — GPT-4o generates realistic EHR narratives; CVAE produces tabular vital-sign distributions; evaluated on fidelity, diversity and privacy risk using SDMetrics. | Python, PyTorch, OpenAI API, SDMetrics, Pandas, Streamlit | MTechPhD IEEE JBHI 2026 |
| 04 | Knowledge Graph Construction and Entity-Relation Extraction from Scientific Papers Using LLM with Graph-RAG Architecture — extracts biomedical entities and relations from PubMed abstracts using GPT-4o structured output; stores knowledge graph in Neo4j; Graph-RAG enables multi-hop reasoning queries. | Python, LangChain, GPT-4o, Neo4j, SpaCy, FastAPI | MTechPhD IEEE Trans. KDE 2026 |
| 05 | LLM-Based Automated Financial Report Generation with Table Understanding, Chart Interpretation and Regulatory Compliance Checking — GPT-4V parses PDF annual reports, extracts tables/charts, generates analyst-grade commentary and flags SEC/SEBI compliance issues using custom guardrail system. | Python, GPT-4V, LangChain, pdfplumber, FastAPI, React | MTech IEEE Access 2026 |
| 06 | Controllable Text Summarisation Using Instruction-Tuned LLM with Length, Tone and Audience Constraints via Structured Prompt Templates — fine-tunes Llama 3.1 8B on CNN/DailyMail with style-conditioned prompts; ROUGE-L, BERTScore and human preference evaluation; deployed as Streamlit web app. | Python, HuggingFace, PEFT LoRA, Streamlit, BERTScore | MTech IEEE Trans. NLP 2026 |
Agentic AI — Multi-Agent Systems
AutoGen · CrewAI · LangGraph · ReAct · Tool-Use · Reflection · Memory · Planning
| # | MTech Python Project Title | Tools | Level & Journal |
|---|---|---|---|
| 07 | Autonomous Multi-Agent Research Paper Summariser Using AutoGen with Planner, Retriever, Critic and Writer Agent Roles — four-agent AutoGen pipeline: Planner decomposes query, Retriever fetches arXiv papers via Semantic Scholar API, Critic evaluates relevance, Writer synthesises final literature review; human-in-the-loop approval before final output. | Python, AutoGen, OpenAI GPT-4o, Semantic Scholar API, FastAPI | MTechPhD IEEE Access 2026 |
| 08 | CrewAI Multi-Agent Supply Chain Risk Intelligence System with Web Search, News Analysis and Risk Scoring Agent Crew — five CrewAI agents collaboratively monitor supplier news, geopolitical risk signals and logistics disruptions; output risk dashboard with mitigation recommendations; tool-use with DuckDuckGo and NewsAPI. | Python, CrewAI, LangChain, OpenAI, NewsAPI, Streamlit | MTech IEEE Trans. Engineering Management 2026 |
| 09 | LangGraph Stateful Agentic Workflow for Autonomous Customer Support with Intent Classification, Tool Calling and Escalation Logic — LangGraph DAG-based agent with persistent conversation memory (Redis), intent routing nodes (billing, technical, returns), tool nodes (CRM lookup, order status API) and automatic human escalation on low-confidence paths. | Python, LangGraph, OpenAI, Redis, FastAPI, Docker | MTech IEEE Access 2026 |
| 10 | ReAct Agent for Autonomous Scientific Hypothesis Generation and Experimental Design Using Tool-Augmented LLM Reasoning Loop — ReAct agent iterates Thought → Action → Observation cycles using PubMed search, Python code execution and web tools; generates testable hypotheses for drug-target interaction problems; self-evaluates with critic LLM. | Python, LangChain ReAct, GPT-4o, PubMed API, Jupyter Kernel | MTechPhD Nature Machine Intelligence 2026 |
| 11 | Self-Reflective AI Agent with Critique-and-Revise Loop for Long-Form Essay Writing with Factual Grounding and Coherence Scoring — generator LLM writes essay draft; reflector LLM critiques factual accuracy and coherence; reviser LLM applies targeted edits; loop runs until quality threshold reached; evaluated on FactScore and coherence metrics. | Python, LangGraph, GPT-4o, FactScore, NLTK, Streamlit | MTech IEEE Trans. AI 2026 |
| 12 | Hierarchical Multi-Agent System for Autonomous E-Commerce Product Listing Generation, SEO Optimisation and Competitor Analysis — Manager agent orchestrates: Image Analyser (GPT-4V), Title/Description Writer, SEO Agent (keyword research via SerpAPI) and Competitor Scout; outputs ready-to-publish Amazon listings with SEO score benchmarks. | Python, AutoGen, GPT-4V, SerpAPI, LangChain, FastAPI | MTech IEEE Access 2026 |
| 13 | Autonomous Data Analysis Agent with Python Code Generation, Execution and Natural Language Explanation Using GPT-4o Code Interpreter — conversational data agent accepts CSV uploads; generates pandas/matplotlib code; executes in sandboxed Jupyter kernel; iteratively fixes errors; explains results in plain language; supports SQL databases. | Python, GPT-4o, LangChain, Jupyter, Pandas, Streamlit | MTech IEEE Trans. KDE 2026 |
LLM Fine-Tuning — LoRA, PEFT & RLHF
LoRA · QLoRA · PEFT · Instruction Tuning · DPO · RLHF · Llama · Mistral · Phi-3
| # | MTech Python Project Title | Tools | Level & Journal |
|---|---|---|---|
| 14 | QLoRA Fine-Tuning of Llama 3.1 8B for Domain-Specific Legal Document Drafting with DPO Alignment on Human Preference Data — 4-bit quantised QLoRA fine-tunes Llama 3.1 8B on 50K Indian legal documents; DPO aligns outputs to lawyer-preferred phrasings; benchmarked on CUAD legal NLP tasks; deployed via vLLM for 4× inference speedup. | Python, PEFT QLoRA, BitsAndBytes, DPO Trainer, vLLM, FastAPI | MTechPhD IEEE Trans. NLP 2026 |
| 15 | Instruction Fine-Tuning Mistral 7B for Medical Diagnosis Assistance with LoRA Adapters and RLHF from Doctor Feedback — LoRA fine-tunes Mistral 7B on MedInstruct dataset; PPO-based RLHF with reward model trained on doctor-rated responses; evaluated on MedQA accuracy; safety guardrails prevent harmful advice. | Python, PEFT LoRA, TRL, PPO, HuggingFace, Streamlit | MTechPhD IEEE JBHI 2026 |
| 16 | Continual Learning Fine-Tuning of Phi-3 Mini with LoRA for Multi-Task Code, Math and Reasoning Without Catastrophic Forgetting — LoRA adapter training with Elastic Weight Consolidation (EWC) on HumanEval, MATH and ARC benchmarks sequentially; progressive task evaluation shows <3% forgetting on prior tasks. | Python, PEFT, HuggingFace, EWC, PyTorch, WandB | MTechPhD IEEE Trans. Neural Networks 2026 |
| 17 | Parameter-Efficient Multilingual NMT Fine-Tuning with LoRA Adapters on NLLB-200 for Low-Resource Indian Language Translation — LoRA adapters added to NLLB-200 3.3B for Hindi, Kannada, Tamil, Telugu; BLEU and COMET scores improve 18% vs zero-shot baseline; <0.5% additional parameters trained. | Python, PEFT LoRA, NLLB-200, SacreBLEU, COMET, HuggingFace | MTech ACM Trans. Asian NLP 2026 |
| 18 | LLM Alignment Using Direct Preference Optimisation (DPO) for Reducing Hallucination and Improving Factuality in Open-Domain QA — DPO trains Llama 3 8B on TruthfulQA preference pairs; evaluated on FactScore, TruthfulQA and HaluEval; systematic ablation of DPO β hyperparameter vs factuality-fluency trade-off. | Python, TRL DPO, Llama 3, PEFT, FactScore, HuggingFace | MTechPhD IEEE Trans. AI 2026 |
RAG — Retrieval-Augmented Generation
ChromaDB · FAISS · Pinecone · Hybrid Retrieval · HyDE · Reranking · Graph-RAG · Self-RAG
| # | MTech Python Project Title | Tools | Level & Journal |
|---|---|---|---|
| 19 | Advanced RAG Pipeline with HyDE, Multi-Query Expansion and Cross-Encoder Reranking for Legal Document Retrieval System — HyDE generates hypothetical answer embeddings for better retrieval; multi-query expansion improves recall; Cohere reranker re-scores top-50 chunks; evaluated on LegalBench; LLM generates cited answers. | Python, LangChain, Cohere Reranker, FAISS, GPT-4o, FastAPI | MTechPhD IEEE Trans. KDE 2026 |
| 20 | Self-RAG: Adaptive Retrieval Decision Making for Open-Domain Question Answering Using Retrieve, Reflect and Critique Tokens — implements Self-RAG architecture where LLM autonomously decides when to retrieve, evaluates retrieved docs for relevance and critiques its own generated segments; benchmarked on PopQA and TriviaQA. | Python, HuggingFace, FAISS, vLLM, PyTorch, RAGAS | MTechPhD IEEE Trans. AI 2026 |
| 21 | Multimodal RAG System for Technical Manual Q&A: Retrieving Text, Diagrams and Tables from Engineering PDFs Using ColPali — ColPali encodes entire PDF pages as images; FAISS retrieves top-k pages; GPT-4V answers questions using both text and visual context; precision@10 comparison vs text-only RAG. | Python, ColPali, GPT-4V, FAISS, LlamaIndex, Streamlit | MTech IEEE Access 2026 |
| 22 | RAG-Powered Personalised Learning Tutor with Adaptive Difficulty Using Student Knowledge Graph and LLM Feedback Loop — student knowledge state tracked in Neo4j graph; RAG retrieves misconception-targeted explanations from NCERT/Khan Academy corpus; GPT-4o generates Socratic dialogue; mastery thresholds trigger topic progression. | Python, LangChain, Neo4j, ChromaDB, GPT-4o, React | MTech IEEE Trans. Learning Tech 2026 |
| 23 | Hybrid BM25 + Dense Vector RAG with Dynamic Chunk Boundary Detection for Code Repository Question Answering — code-aware chunking at AST function boundaries; BM25 for exact-match tokens + CodeBERT dense retrieval; reciprocal rank fusion; GPT-4o answers GitHub Copilot-style questions over private codebases. | Python, CodeBERT, BM25 Okapi, FAISS, LangChain, GPT-4o | MTech IEEE Trans. Software Eng 2026 |
| 24 | Temporal RAG for News Intelligence: Retrieving Time-Sensitive Documents with Recency-Weighted Ranking and LLM Timeline Synthesis — news articles indexed with timestamp decay score; hybrid retrieval with recency weighting; LLM synthesises chronological narrative of events; evaluated on TREC News Track 2025 queries. | Python, LangChain, Pinecone, OpenAI, NewsAPI, Streamlit | MTech IEEE Access 2026 |
Multimodal & Vision AI
GPT-4V · LLaVA · CLIP · BLIP-2 · Video Understanding · Medical Imaging · Visual Q&A
| # | MTech Python Project Title | Tools | Level & Journal |
|---|---|---|---|
| 25 | Multimodal Medical Report Generation from Chest X-Ray Using LLaVA 1.6 Fine-Tuned on MIMIC-CXR with Radiology Terminology Grounding — LoRA fine-tunes LLaVA 1.6 7B on 200K MIMIC-CXR image-report pairs; structured output with findings, impression and recommendations; BLEU/CIDEr/RadGraph F1 evaluation; deployed via FastAPI. | Python, LLaVA, PEFT LoRA, MIMIC-CXR, FastAPI, Gradio | MTechPhD IEEE Trans. Medical Imaging 2026 |
| 26 | CLIP-Based Zero-Shot Defect Classification in Industrial PCB Inspection Using Natural Language Defect Descriptions — CLIP encodes defect images and text descriptions (solder bridge, open pad, missing component); zero-shot accuracy on FICS-PCB dataset; active learning loop for few-shot adaptation; Gradio demo interface. | Python, CLIP, PyTorch, FICS-PCB, scikit-learn, Gradio | MTech IEEE Trans. Industrial Informatics 2026 |
| 27 | Video Understanding Agent: Temporal Grounding and Dense Video Captioning with GPT-4V and BLIP-2 Frame Encoder Pipeline — extracts keyframes at adaptive intervals; BLIP-2 generates per-frame captions; GPT-4o synthesises temporal narrative; supports natural language temporal queries ("what happened at minute 12?"); tested on ActivityNet Captions. | Python, BLIP-2, GPT-4V, OpenCV, LangChain, FastAPI | MTech IEEE TCSVT 2026 |
| 28 | Visual Question Answering for Satellite Remote Sensing Images Using GeoChat Multimodal LLM with Spatial Reasoning Evaluation — GeoChat fine-tuned on RSVQA and RSIVQA datasets; supports count, presence, comparison and spatial-relation questions over Sentinel-2 and DOTA imagery; benchmarked on GeoQA accuracy metrics. | Python, GeoChat, HuggingFace, PEFT, Rasterio, Streamlit | MTechPhD IEEE GRSL 2026 |
| 29 | Multimodal Sentiment Analysis Fusing Text, Image and Audio Using Cross-Modal Attention Transformer on CMU-MOSI and CMU-MOSEI Datasets — cross-modal transformer aligns textual (BERT), visual (ViT) and acoustic (wav2vec 2.0) modalities; late-fusion with learned gate weights; MAE and accuracy benchmarked against unimodal baselines. | Python, PyTorch, BERT, ViT, wav2vec2, CMU-MOSI, FastAPI | MTechPhD IEEE Trans. Affective Computing 2026 |
NLP & Advanced Text Generation
Summarisation · Classification · NER · Information Extraction · Bias Detection · Hallucination
| # | MTech Python Project Title | Tools | Level & Journal |
|---|---|---|---|
| 30 | LLM-Based Hallucination Detection and Factual Consistency Scoring for Abstractive Summarisation Using SummaC and FactScore — GPT-4o summarises news articles; FactScore decomposes claims and verifies each against Wikipedia; SummaC computes NLI-based consistency; hallucination rate benchmarked across 5 LLMs on CNN/DM and XSum. | Python, GPT-4o API, FactScore, SummaC, HuggingFace, Streamlit | MTechPhD IEEE Trans. NLP 2026 |
| 31 | Generative AI for Clinical NER and ICD-10 Code Assignment from Discharge Summaries Using GPT-4o with Few-Shot Chain-of-Thought — few-shot CoT prompts guide GPT-4o to extract diagnoses, procedures and medications from i2b2 clinical notes; maps to ICD-10 codes; F1 benchmarked against BioBERT and Med-NER fine-tuned models. | Python, GPT-4o, SpaCy, i2b2 Dataset, LangChain, FastAPI | MTech IEEE JBHI 2026 |
| 32 | Social Media Misinformation Detection Using LLM-Based Evidence-Augmented Classification with Twitter/X COVID-19 Fact-Check Dataset — LLM retrieves evidence from fact-check sites via RAG; classifies claim as True/False/Unverified with explanation; evaluated on CoAID and LIAR-Plus; human annotation agreement study included. | Python, LangChain, OpenAI, FAISS, Streamlit, scikit-learn | MTech IEEE Trans. Computational Social Systems 2026 |
| 33 | Cross-Lingual Transfer Learning for Low-Resource Indian Language NLP Tasks Using mBERT and XLM-RoBERTa with Prompt-Based Fine-Tuning — prompt-based fine-tuning on NLI, sentiment and NER for Kannada, Telugu, Tamil and Malayalam using IndicGLUE benchmark; comparison of full fine-tune vs LoRA adapter approaches. | Python, XLM-RoBERTa, PEFT, IndicGLUE, HuggingFace Trainer | MTechPhD ACM TALLIP 2026 |
| 34 | LLM-Powered Automated Legal Contract Review with Clause Extraction, Risk Flagging and Plain-Language Explanation — GPT-4o with structured output extracts 24 CUAD clause types; risk rules engine flags non-standard terms; plain-English rewriting for non-lawyer audience; benchmarked on CUAD contract dataset. | Python, GPT-4o Structured Output, LangChain, pdfplumber, FastAPI | MTech IEEE Access 2026 |
AI Code Generation & Dev Agents
Code Completion · Test Generation · Bug Detection · CI/CD Agents · Self-Debugging · CodeReview
| # | MTech Python Project Title | Tools | Level & Journal |
|---|---|---|---|
| 35 | Autonomous LangGraph Coding Agent with Self-Debugging Loop, Unit Test Execution and Iterative Repair on HumanEval and SWE-Bench — LangGraph agent generates Python code, runs unit tests in sandboxed subprocess, inspects test failures, generates repair plan and re-codes; 87% pass@1 on HumanEval; SWE-Bench instance resolution tracking. | Python, LangGraph, GPT-4o, Docker Sandbox, pytest, FastAPI | MTechPhD IEEE Trans. Software Eng 2026 |
| 36 | LLM-Powered Automated Code Review Agent with Security Vulnerability Detection, CWE Classification and Fix Suggestion — GPT-4o reviews pull request diffs; detects SQL injection, XSS, path traversal, hardcoded secrets; classifies to CWE taxonomy; suggests secure code fixes; integrated with GitHub Actions via webhook; benchmarked on CWE-BENCH. | Python, OpenAI, GitHub API, LangChain, FastAPI, Docker | MTech IEEE Trans. Reliability 2026 |
| 37 | Fine-Tuned CodeLlama 13B for Domain-Specific SQL Query Generation from Natural Language Using Spider 2.0 and BIRD Benchmarks — LoRA fine-tunes CodeLlama 13B on Spider 2.0 (2,000+ complex NL-to-SQL pairs); evaluated on execution accuracy and exact match; schema-linking pre-processing step reduces hallucinated column names. | Python, CodeLlama, PEFT LoRA, Spider 2.0, BIRD, SQLite | MTechPhD IEEE Trans. KDE 2026 |
| 38 | Multi-Agent DevOps Pipeline with AutoGen Agents for Requirements-to-Code, Code-to-Test and Test-to-Deploy Workflow Automation — RequirementsAgent decomposes specs; CoderAgent writes implementation; TestAgent generates pytest suite; SecurityAgent scans for CVEs; DeployAgent triggers Docker build and push to ECR; human approval gate before prod. | Python, AutoGen, GPT-4o, Docker, AWS ECR, GitHub Actions | MTech IEEE Access 2026 |
| 39 | AI-Assisted Technical Debt Detection and Refactoring Agent Using LLM Code Smell Analysis on Open-Source Python Repositories — AST-based code smell detector (long methods, god classes, duplicate code) surfaces issues; GPT-4o explains debt and proposes refactored code; developer accepts/rejects via VS Code extension; evaluated on PyPI top-100 repos. | Python, GPT-4o, AST, radon, VS Code Extension API, FastAPI | MTech IEEE Trans. Software Eng 2026 |
Diffusion Models & Image GenAI
Stable Diffusion XL · ControlNet · InstructPix2Pix · LoRA Training · DreamBooth · Video Diffusion
| # | MTech Python Project Title | Tools | Level & Journal |
|---|---|---|---|
| 40 | ControlNet-Guided Medical Image Synthesis for Data Augmentation in Rare Disease Segmentation Using Skin Lesion Conditioning — ControlNet conditioned on lesion masks generates 50K synthetic ISIC dermoscopy images; YOLOv11 segmentation model trained on real + synthetic data achieves 12% higher mDice on rare lesion classes. | Python, ControlNet, Stable Diffusion, PyTorch, YOLOv11, ISIC | MTechPhD IEEE Trans. Medical Imaging 2026 |
| 41 | DreamBooth LoRA Fine-Tuning of SDXL for Personalised Fashion Design Generation with Style Consistency and Textile Pattern Control — DreamBooth LoRA trains SDXL on 15 reference garment images; generates on-model photoshoots across poses and backgrounds; texture-control via ControlNet tile preprocessor; evaluated on CLIP-T and aesthetic scoring. | Python, SDXL, DreamBooth, LoRA, ControlNet, Diffusers, Gradio | MTech IEEE Access 2026 |
| 42 | InstructPix2Pix-Based Satellite Image Change Simulation for Climate Impact Visualisation and Urban Growth Prediction — fine-tunes InstructPix2Pix on paired pre/post-event Sentinel-2 imagery for flood inundation and deforestation simulation; text-guided editing ("show after 10 years of urban expansion"); FID and SSIM evaluation. | Python, InstructPix2Pix, Diffusers, Rasterio, PyTorch, Gradio | MTechPhD IEEE GRSL 2026 |
| 43 | Text-to-3D Molecule Generation Using Equivariant Diffusion Model (EDM) with SMILES Conditioning for Drug Discovery Applications — EDM generates 3D molecular conformations from text-described pharmacophores; valid SMILES rate, drug-likeness (QED) and synthesisability (SA score) evaluated; docking score compared against known actives on ChEMBL targets. | Python, EDM, RDKit, PyTorch Geometric, Streamlit, ChEMBL | MTechPhD Nature Machine Intelligence 2026 |
| 44 | Latent Diffusion Model for High-Resolution Architectural Floor Plan Synthesis from Room Type Annotations with Constraint Satisfaction — LDM conditioned on room-type bounding box annotations generates realistic floor plans; graph neural network post-processor enforces door/window placement constraints; FID, diversity and constraint satisfaction rate reported. | Python, LDM, PyTorch, HuggingFace Diffusers, Matplotlib, Gradio | MTech IEEE Trans. CG & Applications 2026 |
AI Digital Twins & Simulation
LLM-Driven Simulation · Physics-Informed AI · Synthetic Data · Predictive Maintenance · Smart City
| # | MTech Python Project Title | Tools | Level & Journal |
|---|---|---|---|
| 45 | LLM-Driven Conversational Digital Twin for Smart Manufacturing: Natural Language Control, Anomaly Explanation and What-If Simulation — GPT-4o with tool-use queries OPC-UA factory sensors, runs what-if simulations (temperature ±10°C), explains anomalies in plain language; LangGraph agent loops back sensor data to refine predictions. | Python, LangGraph, GPT-4o, OPC-UA, MQTT, FastAPI, React | MTechPhD IEEE Trans. Industrial Informatics 2026 |
| 46 | Physics-Informed Neural Network (PINN) + LLM Digital Twin for Structural Health Monitoring of Bridges with Generative Damage Scenario Simulation — PINN trained on FEM simulation data predicts strain distribution; LLM generates synthetic damage scenarios; combined model forecasts remaining useful life; validated on Z24 Bridge benchmark dataset. | Python, DeepXDE, PyTorch, GPT-4o, FastAPI, Plotly Dashboard | MTechPhD IEEE Sensors Journal 2026 |
| 47 | Generative AI Digital Twin for Smart Grid Energy Management: LLM-Orchestrated Demand Forecasting, Fault Simulation and EV Charging Optimisation — GPT-4o orchestrates LSTM demand forecaster, GAN-based fault scenario generator and RL-based EV charging scheduler; natural language queries return grid state analysis; tested on PJM LOAD dataset. | Python, LangChain, PyTorch LSTM, Stable-Baselines3, Streamlit | MTechPhD IEEE Trans. Smart Grid 2026 |
| 48 | Agentic AI Digital Twin for Personalised Cancer Treatment Planning: Multi-Agent Simulation of Tumour Growth, Drug Response and Side-Effect Prediction — AutoGen agents simulate tumour microenvironment (proliferation/necrosis/angiogenesis), pharmacokinetic drug diffusion and immune response; LLM synthesises oncologist-readable treatment recommendation report. | Python, AutoGen, PyTorch, BioGears API, NumPy, Plotly | MTechPhD IEEE JBHI 2026 |
| 49 | Generative AI-Powered Smart City Digital Twin: Synthetic Traffic Flow Simulation, Pedestrian Density Prediction and LLM Event Impact Analysis — SUMO traffic simulator generates synthetic scenarios; DCGAN augments edge-case traffic images; GPT-4o analyses event-day impact; LangGraph agent recommends traffic diversion routes via HERE Maps API. | Python, SUMO, DCGAN, LangGraph, GPT-4o, HERE Maps API | MTech IEEE Trans. ITS 2026 |
Federated AI & Privacy-Preserving GenAI
Federated LLM · Differential Privacy · Secure Aggregation · PySyft · Confidential AI · On-Device LLM
| # | MTech Python Project Title | Tools | Level & Journal |
|---|---|---|---|
| 50 | Federated LoRA Fine-Tuning of Llama 3 Across Decentralised Hospital Nodes with Differential Privacy and Secure Gradient Aggregation — federated learning using Flower FL framework; LoRA adapters trained locally on each hospital node; Opacus DP-SGD adds (ε,δ)-differential privacy; secure aggregation via Shamir Secret Sharing; benchmarked on i2b2 NER. | Python, Flower FL, PEFT LoRA, Opacus DP, HuggingFace, FastAPI | MTechPhD IEEE Trans. Information Forensics 2026 |
| 51 | On-Device Generative AI with Quantised Phi-3 Mini 3.8B Running on Raspberry Pi 5 Using ONNX Runtime and 4-bit GPTQ Quantisation — GPTQ 4-bit quantises Phi-3 Mini to 2.1 GB; ONNX Runtime int4 inference on Raspberry Pi 5; 6 tokens/second throughput; power consumption profiling; privacy-preserving offline chatbot for healthcare workers. | Python, GPTQ, ONNX Runtime, Phi-3, Raspberry Pi 5, FastAPI | MTech IEEE IoT Journal 2026 |
| 52 | Privacy-Preserving Synthetic EHR Generation Using Differentially-Private GAN (DP-GAN) with LLM-Based Quality Evaluation — CTGAN with DP-SGD (ε=1.0) generates synthetic MIMIC-IV tabular patient records; GPT-4o evaluates clinical plausibility; utility-privacy trade-off curve plotted on ML downstream task performance. | Python, CTGAN, Opacus, SDMetrics, GPT-4o, Pandas, Streamlit | MTechPhD IEEE JBHI 2026 |
| 53 | Membership Inference Attack Evaluation and Watermarking Defense for Large Language Models Fine-Tuned on Proprietary Datasets — implements 5 MIA attack strategies on fine-tuned GPT-2 and Llama 3; evaluates attack success rate vs model utility; LLM watermarking using hard/soft green-list token generation; TPR@1%FPR metric. | Python, HuggingFace, PyTorch, MIA Toolkit, Watermark Library | MTechPhD IEEE Trans. Information Forensics 2026 |
| 54 | Confidential AI Inference Using Intel TDX Trusted Execution Environment for Privacy-Preserving GenAI API Serving — Llama 3.1 8B runs inside Intel TDX TEE; remote attestation verifies computation integrity; encrypted input/output channels prevent cloud provider data access; throughput vs security overhead benchmarked. | Python, Intel TDX SDK, ONNX Runtime, FastAPI, Docker | MTechPhD IEEE Trans. Dependable Systems 2026 |
Extended Topics
Additional GenAI & Agentic AI MTech Python Topics 55–65
Further IEEE 2026 topics across emerging sub-domains — each available with full source code, base paper and report.
| # | Project Title | Tools | Level |
|---|---|---|---|
| 55 | LLM-Powered Autonomous Web Agent Using Playwright with Structured Action Space for Task Completion on WebArena Benchmark | Python, LangGraph, GPT-4o, Playwright, WebArena | MTechPhD |
| 56 | GraphRAG: Microsoft Graph-Based Retrieval-Augmented Generation for Community-Aware Question Answering on Large Document Corpora | Python, GraphRAG (Microsoft), Neo4j, GPT-4o, FastAPI | MTechPhD |
| 57 | Mixture-of-Experts (MoE) LLM Fine-Tuning with Sparse Gating for Multi-Domain Specialisation Using Mistral-8x7B Architecture | Python, Mistral MoE, PEFT, HuggingFace, PyTorch, WandB | PhD |
| 58 | Generative AI for Automated Scientific Literature Gap Analysis and Novel Research Direction Suggestion Using Citation Graph RAG | Python, GPT-4o, Semantic Scholar API, Neo4j, LangChain | MTechPhD |
| 59 | Autonomous AI Product Manager Agent: CrewAI Pipeline for PRD Generation, Competitive Analysis and Sprint Planning from User Interviews | Python, CrewAI, GPT-4o, Notion API, JIRA API, Streamlit | MTech |
| 60 | Text-to-SQL Agentic System with Schema Linking, Ambiguity Resolution and Explanation Generation Using LangGraph Multi-Step Reasoning | Python, LangGraph, GPT-4o, Spider 2.0, PostgreSQL, FastAPI | MTech |
| 61 | Generative AI Audio Synthesis for Personalized Podcast Episode Creation Using Bark TTS and LLM Script Writing Agent | Python, LangChain, Bark TTS, GPT-4o, FastAPI, React | MTech |
| 62 | LLM-Based Automated Penetration Testing Agent with ReAct Loop for CVE Identification, Exploit Planning and Report Generation | Python, LangGraph, GPT-4o, Metasploit API, Nmap, Docker | MTechPhD |
| 63 | Speculative Decoding with Draft Model for 3× LLM Inference Speedup on Consumer GPUs Without Accuracy Loss | Python, HuggingFace, vLLM, PyTorch, NVIDIA A100/RTX 4090 | MTechPhD |
| 64 | Multimodal Agentic AI Travel Planner: CrewAI Agents Using Maps, Weather, Images and LLM for Personalised Itinerary Generation | Python, CrewAI, GPT-4V, Google Maps API, OpenWeather, React | MTech |
| 65 | Constitutional AI and RLHF Pipeline for Building Safe, Helpful and Harmless LLM Assistants Following Anthropic's CAI Framework | Python, TRL, PEFT, HuggingFace, PyTorch, WandB, FastAPI | MTechPhD |
Need a personalised IEEE 2026 MTech Python project in Generative AI or Agentic AI?
Tell us your MTech specialisation (CS / AI / Data Science / IT), preferred domain (GenAI / Agentic AI / LLM / RAG / Diffusion / Digital Twin), university and submission deadline — we'll recommend the best-fit topic, IEEE base paper and start implementation within 24 hours. Full package: Python source code, IEEE base paper, architecture diagram, report, PPT and viva support.
FAQ
MTech Python Projects — Frequently Asked Questions
Top IEEE 2026 MTech Python project topics in Generative AI include: RAG-based enterprise knowledge assistant using LangChain and ChromaDB (GPT-4o), autonomous multi-agent research summariser using AutoGen, QLoRA fine-tuning of Llama 3.1 for domain-specific tasks, LangGraph stateful agentic customer support workflow, Self-RAG adaptive retrieval for open-domain QA, ControlNet medical image synthesis for data augmentation, LLM hallucination detection using FactScore, CLIP zero-shot industrial defect detection, federated LoRA fine-tuning with differential privacy, and LLM-powered digital twin for smart manufacturing. All include IEEE Xplore 2025–2026 base papers and complete Python source code.
Agentic AI MTech projects use: LangChain and LlamaIndex for LLM application development; LangGraph for stateful, cyclic agent workflows with memory and branching; AutoGen (Microsoft) for multi-agent conversations where agents collaborate, critique and iterate; CrewAI for role-based agent crews with defined goals and tools; OpenAI Function Calling and Tool Use for structured agent-tool interaction; Tavily / DuckDuckGo / SerpAPI for web search tools; Playwright / Selenium for browser-use agents; and Docker sandbox for code execution agents with safe Python execution.
Yes. All 65 MTech Python project topics are IEEE 2025–2026 aligned and formatted for VTU Bangalore, Anna University Chennai, JNTU Hyderabad/Kakinada, Amrita, PES, RV, BMS, Manipal and all autonomous university requirements. Each project includes: IEEE Xplore base paper with DOI, complete Python source code with documentation, university-format project report (Chapter 1–5 with abstract, literature review, architecture, implementation, results and conclusion), system architecture and data flow diagrams, PPT presentation with screenshots and results, and viva Q&A preparation covering LLM, RAG, agents, prompt engineering, vector databases, fine-tuning and diffusion model concepts.
Most GenAI MTech projects can be developed using: (1) Cloud APIs — OpenAI GPT-4o, Gemini 1.5 Pro, Anthropic Claude via API (no local GPU needed); (2) Google Colab Pro / Kaggle — free T4 GPU for HuggingFace model inference and LoRA fine-tuning of 7B models; (3) Local PC with 16GB RAM — sufficient for running quantised 4-bit models (Llama 3.1 8B via Ollama), LangChain/AutoGen agent development and Streamlit demos; (4) GPU server (A100/RTX 4090) — required for full fine-tuning of 13B+ models or Stable Diffusion training. We guide students on cost-optimised setups and provide pre-trained model weights where applicable.
Project Lab
Python AI Project Lab — Bangalore
GPU workstations for LLM fine-tuning, LangChain/AutoGen agentic pipeline development environments, Stable Diffusion image generation stations and multi-agent system demo rigs for MTech students.
LLM Fine-Tuning GPU Station
Agentic AI Development Lab
RAG Pipeline Workstation
Diffusion Model Station