Headline

Building Gomarble.ai || MTS-1 NLP Developer @Alphastream.ai ll FORMER SWE INTERN @Bytelearn.ai || Data Science ll COMPETITIVE PROGRAMMER || IET’23

Location

Bengaluru

AI stack

aistudio

anthropic apis

antigravity

claude code

claude

codex

Work experience

Data Scientist at GoMarble, Mar 2025 - Present · 1 yr 9 mos, Bengaluru, Karnataka, India

MTS1 NLP Developer at alphastream.ai, Jun 2023 - Mar 2025 · 1 yr 10 mos, Bengaluru, Karnataka, India, • Improved NER classification for financial tables using BAAI/bge-large-en model, boosting accuracy from 87% to 94%. • Developed a RAG pipeline for extracting 50 complex data points with 80% accuracy from 300+ page credit agreement documents, using Elastic Search, Llama Index, and Qwen1.5-72B model. • Implemented and evaluated various ranking profiles, including hybrid profiles, using Vespa. • Created a parallelized system to extract 500 data points in 1.5 minutes from private company profiles. • Built a RAG system to extract direct and calculated data points from financial documents, handling currency and quantum.

Member of Technical Staff at alphastream.ai, Jun 2023 - Mar 2025 · 1 yr 10 mos, Alphastream builds AI document-intelligence products for fixed-income and credit-market analysis. Engineered high-precision RAG pipelines extracting structured data from 300+ page credit agreements using LlamaIndex, ElasticSearch, and Qwen1.5-72B. Key results: ~80% extraction accuracy on a 50-field schema; NER accuracy raised 87% → 94% via optimized BAAI/bge-large-en embeddings; Text-to-SQL engine (Claude 3.5 Sonnet) at 87% execution accuracy on financial-database benchmarks. Designed a parallelized extraction engine pulling 500+ data points from private-company profiles in under 90 seconds; implemented hybrid ranking profiles in Vespa; benchmarked Google Model Garden deployments.

Software Engineer Intern at ByteLearn.com, Jul 2022 - Dec 2022 · 6 mos, Noida, Uttar Pradesh, India, • Worked with solver team: Assisted in creating and debugging 10+ problem templates for student-chatbot interaction . • Developed Python scripts for the chatbot utilising the Low Code Platform, Python, JSON, and Jinja. Used Postman to generate Chatbot Payload after implementing solution code for various problem categories

Education

Bachelor of Technology - BTech, Computer Science and Engineering, Institute of Engineering and Technology

Intermediate, kendriya vidyalaya

Public links

github: https://github.com/akashgomarble

codeforces: https://codeforces.com/profile/coder2k20