# Ashish Solanki

Ashish Solanki is a remote worker on Himalayas, Assistant ML Engineer at Capsitech, based in India.

Source: https://himalayas.app/@ashishsolanki1

```text
AI Engineer building production Document AI, RAG, and LLM/VLM extraction systems that work reliably.
```

| Detail | Value |
| --- | --- |
| Availability | Looking for a job |
| Location | India |
| Authorized to work in | India |
| Salary expectations | 900k-12000k USD |
| Job categories | AI Engineer, Machine Learning Engineer, AI Engineer, AI Consultant, AI Product Manager, Applied AI Engineer, Generative AI Engineer |
| Skills | Lang Chain, LLM, Python, Git, Linux, LlamaIndex, LangGraph, RAG, Embeddings, Vector Search, Hybrid Search, Prompt Engineering, Coordinate Based Parsing, Row Column Reconstruction, OpenCV, YOLO, Azure Document Intelligence, Qwen VL, PyTorch, Hugging Face, Transformers, VLLM, fastAPI, Pydantic, REST APIs, MongoDB, MongoDB Atlas, Azure Blob Storage, Docker, Docker Compose, Background Execution, Callbacks, Structured Logging Systems, Error Handling, GPU OOM Debugging, Selenium, Streamlit, Column, GitHub, JSON, Movement, Postman, Qwen, Ubuntu, Mediapipe |
| LinkedIn | https://https://www.linkedin.com/in/aashish-solanki/ |
| GitHub | https://https://github.com/ashishsingh2210 |

## What I'm looking for

```text
I’m looking for a production-focused AI/Document AI role where I can improve extraction accuracy, build RAG pipelines, and own FastAPI services end-to-end—from dataset and evaluation to deployment and reliability.
```

## About

```text
I’m an AI Engineer focused on production Document AI, RAG, and LLM/VLM-based extraction systems. I build pipelines that reliably handle digital PDFs, scanned PDFs, images, and multi-page files—then I validate and harden them against real edge cases.

At Capsitech, I improved a bank statement extraction system from ~60% to ~95% quality by replacing Azure Document Intelligence fallback with a fine-tuned Qwen-VL-based workflow. I also expanded bank coverage from 12 live/UAT banks to ~60 banks and 138 versions by designing bank/version-specific logic using coordinate parsing, OpenCV line detection, IoU mapping, NMS-style boundaries, and row-merging rules.

I built an HMRC-style RAG chatbot and ingestion pipeline using LangChain and MongoDB Atlas hybrid retrieval, including a recursive, resumable crawler that generated structured datasets for indexing. I improved internal RAG answer quality from ~50% to ~80% across simple, complex, follow-up, keyword-based, indirect, calculation-based, and year-specific queries.

I delivered invoice extraction workflows using YOLO-based field detection plus OCR-to-detection mapping, and then upgraded unknown-template extraction with Qwen-VL (from Qwen-VL-2.5-7B to Qwen-VL-3-4B), raising quality from ~80% to ~90%. In production FastAPI services, I reduced representative scanned/VLM extraction time from ~86 seconds to ~20 seconds with vLLM batching, and strengthened reliability with Pydantic validation, structured logging, error capture, and Docker-based deployment/debugging. I was also recognized with a Best Paper Award (TIACOMP 2024) for emerging technologies and smart systems.
```

## Experience

### Assistant ML Engineer (current)

Capsitech. Sep 2024 - Present

```text
Improved bank statement extraction quality from ~60% to 95% by replacing Azure Document Intelligence fallback with a fine-tuned Qwen-VL VLM extraction workflow, and expanded coverage from 12 to ~60 banks and 138 versions. Built HMRC-style RAG chatbot/data ingestion (improved answers ~50% to 80%) and accelerated production FastAPI document processing from ~86s to 20s using vLLM batching.
```

Skills: Qwen-VL, Azure-Document-Intelligence, RAG, MongoDB, fastAPI, OpenCV, YOLO, OCR-and-PDF-Extraction, VLLM

### AI/ML Engineer Intern

Websoham. Mar 2024 - May 2024

```text
Developed automated bank statement PDF extraction handling encrypted PDFs, table detection/parsing, and multi-page Excel output generation. Worked on video transcription and shorts generation (timestamp mapping, captions, hook selection) and explored OpenCV/MediaPipe-based lip movement analysis for speaker detection.
```

Skills: Encrypted-PDF-Handling, PDF-Parsing, OpenCV, MediaPipe, MoviePy, Mediapipe

## Education

### Chandigarh University

Master of Engineering in Artificial Intelligence, Artificial Intelligence. 2022 - 2024

Activities and societies: Thesis/papers: High-Resolution Fashion Image Generation using Quantum-GAN; Review paper: Comprehensive Analysis on Image Generation using Quantum GAN.

```text
Master of Engineering in Artificial Intelligence at Chandigarh University. Thesis and research focused on high-resolution fashion image generation using Quantum-GAN.
```

### Indian Institute of Business Management

Post Graduate Program in Data Science, Data Science. 2020 - 2021

Activities and societies: 6-month internship/training covering Machine Learning, AI, and Predictive Modeling.

```text
Post Graduate Program in Data Science at the Indian Institute of Business Management. Completed a hands-on 6-month training in machine learning, AI, and predictive modeling.
```

### U.V. Patel College of Engineering

Bachelor of Technology in Mechatronics Engineering, Mechatronics Engineering. 2016 - 2020

Activities and societies: Major project: Automatic conveyor belt for loading/unloading with inspection using pick and place robot.

```text
Bachelor of Technology in Mechatronics Engineering at U.V. Patel College of Engineering. Completed a major project related to an automatic conveyor belt with inspection using a pick-and-place robot.
```

## Tech stack

- Postman
- GitHub
- Docker Compose
- MongoDB
- Gmail
- OpenCV
- Python
- JSON
- PyTorch
- Streamlit
- FastAPI
- MongoDB Atlas
- Ubuntu
- Linux
- Docker
- Azure Blob Storage
- Hugging Face
- LangChain
- LlamaIndex
- Pydantic
- vLLM
- LangGraph
- Qwen
- Column
- Task
- MediaPipe

To contact Ashish, message them from this profile with a Himalayas employer account: [Hire on Himalayas](https://himalayas.app/hire).

## People also viewed

- [Omkar Shinde](https://himalayas.app/@omkarshinde): Software Engineer, CGI
- [Shiv Sunder](https://himalayas.app/@shivsunder): Machine Learning Intern, Quantiphi
- [Sumanth User](https://himalayas.app/@sumanthuser): AI Engineer, UNIPREP
- [Rishabh User](https://himalayas.app/@rishabhuser3): AI Developer, Wizcoder
- [Samarth Singh](https://himalayas.app/@samarthsingh): AI/ML Applications Intern, AmberFlux EdgeAI Private Limited
- [poojitha paladugu](https://himalayas.app/@poojithapaladugu): AI/ML Engineer, Alphastream
- [Jaydatta Bade](https://himalayas.app/@jaydattabade): AI Engineer Individual Contributor, FilersKeepers/Lawstronaut
- [Anjali User](https://himalayas.app/@anjaliuser23): AI / ML Engineer, IndiaLends
- [Deviprasad Sunny](https://himalayas.app/@deviprasadsunny): AI Intern – TechSaksham, Edunet Foundation
- [Ritu Sarsahay](https://himalayas.app/@ritusarsahay): Machine Learning Engineer, Capsitech
