Pune · IST
Portfolio · 2026 edition 18.52° N, 73.86° E Pune, India

AI & ML Engineer

Aashish Bagmar

AI & ML Engineer building production-ready RAG systems, Voice AI applications, intelligent backend platforms, and real-world AI products.

Available for AI/ML Engineering Opportunities

6 Projects
1 Research Publication
1 Production Application

Achievements

Things with a date or a number attached.

  1. Published research paper in IJCA

    Peer-reviewed · Mar 2026

    Research
  2. Production Flutter POS system used by a real business

    Runwalsoft · 80% faster billing

    Production
  3. Promoted from intern to full-time engineer in 5 months

    Runwalsoft · Nov 2025 to Apr 2026

    New
  4. VoiceHack 2026 participant

    Built CareCaller AI healthcare voice agent

    Hackathon
  5. Meta Scaler participant

    Hackathon builder · AI product development

    Hackathon
  6. 10 SGPA in Semesters 7 & 8

    AI & Data Science · APCoER Pune

    Academics
  7. Self-corrective RAG assistant with 74 passing tests

    LangGraph · Express Analytics assignment

    Testing

Experience

Five months from intern to full-time engineer at Runwalsoft, plus a point-of-sale app that a business runs its daily billing on.

Apr 2026 – Present Nashik, India Current role

Junior Software Engineer

Runwalsoft · Nashik, India

80% Faster Billing 5 Staff Daily Offline-First Real-Time Sync 200 Bills in Early Use
  • Built and deployed a production Android POS app in Flutter and Firebase Firestore for a business, now used daily by 5 staff for billing, inventory, and reporting. Billing time per customer dropped 80%, from 2 minutes to 25 seconds.
  • Made the app work without internet: Hive caching and an offline queue store sales locally and sync to Firestore on reconnect.
  • Added Bluetooth ESC/POS receipt printing, admin and staff roles, device locking, and an analytics screen with CSV and Excel export.
  • Released as a direct APK install, with 200 bills processed in early use. Used Claude Code for feature development and code review.
Measured on the live app, 5 staff
  • Flutter
  • Dart
  • Firebase Firestore
  • Hive
  • Provider
  • Claude Code
Nov 2025 – Apr 2026 Nashik, India

Software Engineer Intern

Runwalsoft · Nashik, India

  • Shipped features and fixes for the financial and accounting workflows of a live Flutter and Firebase Firestore app, through Git and senior code review.
  • Promoted from intern to full-time Junior Software Engineer after 5 months.
  • Flutter
  • Firebase Firestore
  • Git

Featured Projects

Six builds, from a RAG pipeline that grades its own retrieval to a Windows app that turns taps on a desk into shortcuts.

Document intelligence · LangGraph

Express Analytics RAG Assistant

Self-corrective RAG pipeline with 74 passing tests — production-grade document intelligence.

LangGraph-orchestrated technical documentation assistant with document grading, bounded retry loops, citation-backed responses, and incremental indexing across PDF, DOCX, MD, and TXT corpora.

An LLM grades every retrieved chunk. When retrieval is weak, the query is rewritten and the search runs again. Uploads are split with semantic chunking and embedded with all-MiniLM-L6-v2, and the Streamlit view shows a confidence score, retry count, and source for each answer.

  • 74 Tests Passed
  • LangGraph Workflow
  • FastAPI Backend
  • LangGraph
  • FastAPI
  • ChromaDB
  • Ollama
  • Streamlit
  • Python
GitHub Live demo & architecture diagram coming soon

Voice + chat agent · Gemini

Nova Multilingual Real Estate Sales Agent

One prompt drives both chat and voice, in English, Hindi, and Hinglish.

A sales agent built with FastAPI and Google Gemini, using browser speech-to-text and Gemini text-to-speech. Answers are restricted to a fixed property knowledge base, and in testing it declined every question outside that base instead of making up an answer.

It covers lead qualification, objection handling, and simulated bookings that recover when a slot fails. Sessions can be paused and resumed, a pytest suite covers booking logic and conversation scenarios, and a dashboard shows aggregate insights across sessions.

  • EN · HI · Hinglish
  • Chat + Voice
  • Declines Off-KB Questions
  • Python
  • FastAPI
  • Google Gemini
  • Gemini TTS
  • Web Speech API
  • pytest

Healthcare voice AI · VoiceHack 2026

CareCaller AI

Real-time healthcare voice agent with clinical risk scoring — built for VoiceHack 2026.

End-to-end conversational AI platform conducting patient check-ins via Whisper STT and TTS, extracting medical adherence data through a 14-question compliance state machine, and predicting LOW / MEDIUM / HIGH clinical risk in real time.

  • Voice AI
  • Real-time Risk Scoring
  • Healthcare AI
  • Python
  • FastAPI
  • Whisper
  • XGBoost
  • Streamlit
  • scikit-learn
GitHub Live demo & architecture coming soon

Offline RAG · Llama 3

DocMind — Offline RAG

Fully offline PDF Q&A — no API keys or internet required after setup.

Multi-document knowledge base using LangChain, Llama 3 via Ollama, and FAISS with disk-cached embeddings, conversational memory, and source chunk visibility through Streamlit and CLI interfaces.

  • Offline AI
  • Multi-PDF QA
  • FAISS Retrieval
  • LangChain
  • FAISS
  • Ollama
  • Llama 3
  • Streamlit
GitHub Live demo & architecture coming soon

Desktop utility · C# / WPF

Holo for Windows

Acoustic tap-zone classification turning desk space into four assignable action triggers.

Windows port of the macOS Holo utility — reimplements HoloCore in C# with WASAPI audio capture, WPF calibration UI, and local action dispatch for shell commands, clipboard, speech, and screenshots.

  • Acoustic ML
  • WPF Desktop
  • 4 Tap Zones
  • C#
  • .NET 8
  • WPF
  • WASAPI
  • NAudio
GitHub Live demo & architecture coming soon

Published research · ML + NLP

AI Stock Analyst

Peer-reviewed StockSense implementation — hybrid ML + BERT sentiment for BUY / SELL / HOLD signals.

Streamlit application combining Linear Regression, Random Forest, and LSTM ensemble with BERT news sentiment analysis, live Yahoo Finance data, and Bloomberg-style pipeline visualizations.

  • Published Research
  • LSTM Ensemble
  • BERT Sentiment
  • Python
  • TensorFlow
  • PyTorch
  • BERT
  • Streamlit

Skills

The tools behind the work above, grouped by where they show up.

AI Engineering14

  • Python
  • LangGraph
  • LangChain
  • RAG
  • LLMs
  • TensorFlow
  • PyTorch
  • ChromaDB
  • Ollama
  • Google Gemini API
  • Sentence Transformers
  • LSTM
  • Sentiment analysis
  • Prompt design

Backend Engineering05

  • FastAPI
  • REST APIs
  • PostgreSQL
  • Microservices
  • pytest

Frontend & Mobile07

  • React
  • Flutter
  • Streamlit
  • TypeScript
  • Firebase Firestore
  • Hive
  • Provider

Cloud & DevOps04

  • Docker
  • AWS
  • Git
  • Firebase

Languages03

  • Python
  • Dart
  • SQL

Tools02

  • Git
  • Claude Code

Research &Publications

One paper, written with three co-authors. I built the full front end and back end of the system it describes.

Education

Anantrao Pawar College of Engineering and Research

Bachelor of Engineering — Artificial Intelligence and Data Science

Aug 2022 – May 2026 · Pune, India

  • Machine Learning
  • Deep Learning
  • Data Structures
  • Database Systems
  • Software Engineering
CGPA
8.68
Semester 7 SGPA
10.0
Semester 8 SGPA
10.0

Ashoka Universal School

Higher Secondary Education (HSC)

Jun 2020 – Jun 2022 · Nashik, India

Certifications

AWS Prompt Engineering

AWS Training Online

Data Analytics Job Simulation

Deloitte Australia · Forage

Let’s WorkTogether

Open to internships, full-time opportunities, freelance projects, and AI collaborations.

bagmaraashish@gmail.com

It's IST in Pune right now.