Agentic LLM Systems
Multi-agent pipelines with CrewAI — role design, prompt engineering, research automation and signal validation. Proven in fintech: 40% less manual research time.
AYUSHYADAV
AI/ML Engineer building agentic LLM systems, RAG pipelines and computer vision that ship.
From CrewAI research agents to YOLOv11 tracking to ECG deep learning — I build across the full AI stack and care about the numbers proving it works.
Scroll
01
About
I'm Ayush Yadav — an AI/ML engineer and data scientist finishing my B.Tech in Electronics & Telecommunication at MITS Gwalior. I build agentic LLM systems, retrieval pipelines, computer vision and deep learning tools.
I freelance as an AI automation consultant, helping small businesses replace manual busywork with n8n pipelines, RAG assistants and self-running dashboards. Previously, I built multi-agent CrewAI systems at CrowdWisdomTrading and Power Platform automations at HOSHŌ DIGITAL.
My work includes a RAG query system handling 1k+ queries/day, a YOLOv11 + DeepSORT tracker with 95%+ post-occlusion ID recovery, and ECG classification at 95.5% accuracy. I'm also Google-certified and a national hackathon finalist.
Currently taking on new small-business automation clients — if your team is drowning in repetitive work, that's exactly what I fix.
40%
Research time cut with CrewAI agents
16+
Projects built end-to-end
1,000+
Queries/day served by my RAG system
95.5%
ECG classification accuracy achieved
02
What I Do
RAG & Search
FastAPI retrieval systems with vector embeddings and LLM inference — modular multi-stage pipelines serving 1k+ queries/day at low latency.
Computer Vision
Real-time detection, tracking and re-identification with YOLOv11, DeepSORT and OpenCV — plus adaptive logic that cut identity mismatches by 30%.
Data & Automation
Small-business automation that pays for itself — n8n workflows, Power Platform apps and Power BI dashboards replacing spreadsheet drudgery with pipelines that just run.
03
Capabilities
ALLM & GenAI
RAGCrewAILangChain
Vector SearchPrompt EngineeringLLM Orchestration
Multimodal RAGEmbeddingsOllama
BMachine Learning & CV
PyTorchTensorFlowScikit-learn
XGBoostOpenCVYOLOv11
DeepSORTSHAPOptuna
CData & MLOps
PythonSQLPandas
NumPyMLflowAirflow
PrefectDockerGCPBigQuery
DBackend & Interfaces
FastAPI.NET 8 / C#REST APIs
WebhooksPower BIPower Apps
StreamlitReactn8n
04
Selected Work
LLM-powered document Q&A: upload any document, ask natural-language questions. Semantic search over vector embeddings feeds grounded LLM answers, served over FastAPI at low latency.
GitHub ↗
GitHub ↗
Engineering notes
- Handles 1k+ queries/day with consistently low latency.
- Modular 4-stage pipeline: embedding → semantic search → LLM inference → synthesis.
- Each stage upgrades independently — no lock-in on any model or index.
Multi-agent AI pipelines built at CrowdWisdomTrading for automated market-intelligence research — agent crews that gather, validate and synthesize trading signals from live sources.
Details on LinkedIn ↗
Details on LinkedIn ↗
Engineering notes
- Cut manual research time by 40% across the research workflow.
- RAG + LLM orchestration validated signals across 3+ use cases.
- Prompt engineering + inference tuning for real-time trading insights.
Real-time multi-object tracking for sports analytics: YOLOv11 detects players, DeepSORT maintains identities through occlusions — enabling movement analysis across game scenarios.
GitHub ↗
GitHub ↗
Engineering notes
- 95%+ ID recovery accuracy after full occlusion events.
- Adaptive bounding-box logic cut identity mismatches by 30%.
- Tuned tracking consistency across camera angles and crowd scenes.
Cardiac arrhythmia detection from ECG signals on the MIT-BIH Arrhythmia dataset (100k+ samples) — preprocessing, wavelet feature extraction and beat classification into diagnostic categories.
GitHub ↗
GitHub ↗
Engineering notes
- 95.5% accuracy, validated by precision, recall and F1 — not just accuracy.
- Wavelet transforms gave noise-robust features from raw ECG.
- Companion deep-learning track: ECG-Classification (TensorFlow/Keras).
Decentralized medical-records platform giving patients true ownership of their data — wallet-based auth, encrypted record storage, and permission-based sharing with healthcare providers.
GitHub ↗
GitHub ↗
Engineering notes
- Solana Wallet Adapter auth — no passwords, keys are identity.
- Encrypted off-chain storage with on-chain access control.
- TypeScript + React + Web3.js end-to-end.
AI health guidance for underserved communities — multilingual support, voice input and remote doctor consultation, powered by Google Gemini. Built for the Google Developer Group Solution Challenge.
GitHub ↗
GitHub ↗
Engineering notes
- Represented team at GDG Solution Challenge 2025.
- Voice-first UX designed for low-literacy and rural users.
- Multilingual responses lower the access barrier to health info.
05
Resume
Experience
2026 — Present
AI Automation Freelancer
Independent · Small-Business Clients
Helping small businesses automate the busywork — LLM-powered workflows, RAG assistants and n8n/Power Platform pipelines that cut manual tasks so owners can focus on the business, not the ops.
Jan 2026 — Apr 2026
AI Intern (CrewAI / LLM)
CrowdWisdomTrading
Designed multi-agent CrewAI pipelines for market intelligence, cutting manual research time by 40%. Built RAG pipelines and LLM orchestration validating signals across 3+ fintech use cases.
Jul 2025 — Jan 2026
Jr. Technical Consultant — Intern
HOSHŌ DIGITAL · Indore
Engineered automated data pipelines with Power Apps and .NET, cutting manual reporting by 30%. Built Power BI dashboards for 2+ business units and delivered solutions for 3 client accounts.
Education & Credentials
2022 — 2026
B.Tech — Electronics & Telecommunication
Madhav Institute of Technology & Science, Gwalior
Certifications
Google Data Analytics Specialization
Google Cloud: Prompt Design in Vertex AI · Multimodal RAG · Gemini & Imagen
Achievements
Finalist — Hacksagon 2025 (ABV IIITM Gwalior)
GDG Solution Challenge 2025 · NPTEL Elite (IIT Kanpur)
Publications
Deep Learning for Arrhythmia Detection Using ECG Signals
A comparative review
06 — Get in touch
+91 83055 45893 · Indore, India — freelance AI automation for small business, and open to full-time AI/ML roles.