Flagship · Multi-AgentLIVE
KharchaBOT AI
Telegram-native multi-agent financial co-pilot. Tracks income, expenses & investments via NL and receipt OCR. 5-model dynamic Gemini fallback engine.
PythonLangGraphFastAPIChromaDBPostgreSQLGemini API
Available for AI/ML roles & internships
Hi, I'm Ankit Ambasta.
AI/ML Engineer building agentic AI systems, real-time data pipelines, and intelligent IoT solutions.
Vadodara,GujaratVIT Vellore @26CSE (Iot)
Agentic AI · RAG SystemsIoT + ML · Edge InferenceLLM Pipelines · Data Eng
About
I was born in Pune and grew up across Kashmir, Agra, West Bengal, and Gujarat - adapting fast and seeing problems from multiple angles. I didn't just study computing; I became obsessed with building systems that actually run.
I specialize in LLM-powered agentic pipelines (LangGraph, RAG, multi-agent orchestration) and IoT-ML hybrids (ESP32, MQTT, TFLite Micro). I deploy, benchmark, and iterate - not just prototype.
8.7
CGPA
4+
Projects
1
Paper
Selected Work
Offline RAG
GridMind
GridMind is an offline AI-powered survival assistant that can answer critical survival questions when internet or cloud AI is unavailable
PythonFAISSOllamaLangChain
IoT · Digital Twin
VoltGuard
IoT battery telemetry digital twin. ESP32 + INA219 streamed over MQTT, XGBoost + LSTM on NASA dataset. Real-time predictive health monitoring.
ESP32MQTTXGBoostLSTM
Web App · Finance
KharchaWise
Scalable real-time financial management web app. 500+ concurrent users, comprehensive analytics dashboard, live Firebase sync.
JavaScriptFirebaseREST APIs
I build real production systems - not academic demos. From multi-agent AI co-pilots to offline RAG engines and IoT digital twins, each project is designed for real users in real conditions.
Skills & Tools
Languages
PythonSQLC++JavaScriptJava
AI / LLM Stack
LangChainLangGraphRAGFastAPIChromaDBFAISSPrompt Engineering
ML & Deep Learning
TensorFlowKerasXGBoostsklearnCNN / LSTMPandasPySpark
IoT & Edge
ESP32MQTTTFLite MicroRaspberry PiI2C / SPI
Databases
PostgreSQLFirebaseMongoDBRedisSQLite
Cloud & DevOps
GitLinuxGCPBigQueryRenderVercel
Research
Preprint - Zenodo - 2025
Non-Intrusive Load Monitoring Using Dual-Branch CNN with Attention Mechanisms and XGBoost Classifier
Appliance-level energy disaggregation from smart meter readings - no hardware modification required. Dual-branch CNN extracts spatial and temporal features, paired with XGBoost for appliance identification. Benchmarked on REFIT dataset.
0.977
F1 · Microwave
0.923
F1 · Washing Machine
REFIT
Benchmark Dataset
Certifications & Achievements
Top 3 - Internal Hackathon
Wearable smart suit prototype for factory workers - real-time biometric monitoring, hazard detection using ESP32.
Education
- Jul 2022 - Jul 2026
B.Tech - CSE with IoT
Vellore Institute of Technology (VIT)CGPA: 8.7. DSA, System Design, DBMS, AI, ML, Cloud Computing, IoT specialization.
- 12th Grade (CBSE)
Higher Secondary - 92.2%
Air Force School AgraPhysics, Chemistry, Maths, Computer Science. Top 10 in school.
- 10th Grade (CBSE)
Secondary Education - 87.2%
Air Force School AgraStrong science and mathematics foundation.
Experience
- Dec 2025 - Jan 2026
Data Science Trainee
Wipro CoE - VIT Vellore60-hour intensive program. Pandas, NumPy, SQL, PySpark, Snowflake, data warehousing. QLoRA fine-tuned BERT for financial sentiment classification.
- Leadership
Youth Red Cross Volunteer
Community ServiceLed 20-person team across health awareness campaigns, blood donation drives, and social welfare initiatives.
Say Hii
Get in Touch
Looking for AI/ML roles or internships. I build real production systems - if you need someone who can own a problem end-to-end, let's talk.