Sankalp Shubham

Co-Founder

Bengaluru, Karnataka, India3 yrs 2 mos experience

AI EnabledAI ML Practitioner

Key Highlights

Expert in Large Language Models and AI engineering.
Proven track record in developing AI-driven solutions.
Strong foundation in deep learning and model implementation.

Stackforce AI infers this person is a SaaS-focused AI engineer with expertise in LLMs and backend development.

Contact

Skills

Core Skills

Large Language Models (llm)Llm EvaluationAgentsRetrieval-augmented Generation (rag)Ruby On RailsJavaPython (programming Language)Deep LearningNatural Language Processing (nlp)

Other Skills

Distributed SystemsPrompt EngineeringVoice agentsSystems DesignGenerative AIRubySpring BootRedisC++PyTorchLong Short-term Memory (LSTM)Gated Recurrent Units (GRU)TransformersAlgorithmsWriting

About

Founding AI engineer at Nevara and interested in evals, post-training, upcoming trends Check out my Featured blogs here https://sankalp.bearblog.dev/featured/ Check out my latest blog - https://sankalp.bearblog.dev/how-prompt-caching-works/ Github: https://github.com/sankalp1999 <to_be_updated_below> I was working at US based fintech company Zuora(NYSE: ZUO) as a backend engineer for the past ~2 years. My production experience includes 2 years of backend/generalist software engineering in production environments with varying techstacks (Java, Ruby). Have been venturing into AI consulting/working as contractor since July. Helped consult a couple of startups and worked on RAG prototypes for a founder on extremely short notice. Working on a agentic codegen project (automating creation of recruitment programming assignments) since Sept 2024. I have built projects utilising GenAI workflows (RAG based search, agents, vectorDBs, LLMs). Other than applied ML, I am also familiar with the foundational/training layer. I possess a strong foundation in deep learning, with experience implementing research papers and building models using PyTorch (e.g., CaptionBot v2.0 based on "Show, Attend and Tell"). You can check out some of my technical writing at https://sankalp.bearblog.dev

Experience

3 yrs 2 mos

Total Experience

9 mos

Average Tenure

9 mos

Current Experience

Nevara

2 roles

Founding AI Engineer

Promoted

Aug 2025 – Present · 8 mos · Remote

Crafting the AI features in Nevara - an AI sales assistant meant for Account Executives

Large Language Models (LLM)LLM evaluationPython (Programming Language)Distributed Systems

Consulting AI Engineer

May 2025 – Present · 11 mos · Remote

Improved AI sales coach roleplay personas by iterating on prompts with advanced techniques, upgrading to SOTA models, and optimizing context workflows for better performance
Extracted out a spec and wrote basic evals for persona quality and developed a tool to align LLM as a judge outputs with domain expert's judgement output
Deployed new persona voice agents for enterprise customer under tight timeline
Led and mentored 2 MS CS interns to build LLM-based recommendation engine for enterprise customer

LLM evaluationLarge Language Models (LLM)Prompt EngineeringVoice agents

Antiwork

Open Source Developer

Jun 2025 – Aug 2025 · 2 mos · New York, United States · Remote

Contributed to Gumroad, Helper and Gumboard codebases earning bounties 10K+ USD
More details here (WIP blog)
https://sankalp.bearblog.dev/antiwork-pr-contribution/

Lancedb

Independent Consultant / Technical Writer

Nov 2024 – Dec 2024 · 1 mo · Remote

Wrote the following 2 articles read by thousands of people
https://blog.lancedb.com/rag-codebase-1/
https://blog.lancedb.com/building-rag-on-codebases-part-2/
It's 2025 and these are still the most detailed blogs you can find on codebase indexing on the internet

Large Language Models (LLM)

Scaled focus

Applied AI/Product Engineer

Sep 2024 – Feb 2025 · 5 mos · Bangalore Urban, Karnataka, India · Remote

Project: Automated Generation of SWE Hiring Assignments using LLMs (Sep–Dec 2024)
Collaborators: Nirant Kasliwal, Rajaswa Patil
Designed and implemented a configurable, stack-agnostic pipeline that automatically generated programming assignment boilerplates, significantly reducing manual effort (saving thousands of developer-hours monthly) for client Iamneo.
Integrated and adapted `swe-agent` into our full-stack, agentic code-generation pipeline.
Conducted detailed technical evaluations of various agentic code-generation tools (`SWE-agent`, `OpenHands`) to identify the optimal solutions for our workflow.
Project: ragchallenges.com (Dec 2024–Jan 2025; paused due to funding constraints)
Built an evaluation platform enabling users to submit outputs from RAG/agentic pipelines for benchmarking against domain-specific datasets (Legal, Finance, Healthcare, Protein). Provided metrics including Faithfulness, Context Recall, Multi-Turn Performance, and Aspect Critique.
Developed the initial full-stack prototype (NextJS frontend + FastAPI backend) independently from scratch.
Implemented an asynchronous evaluation system utilizing Redis and Celery.
Gained insights into LLM evaluation loops, iterative product improvement, and client-driven feedback.
Project: Financial Data Ingestion and Reconciliation Automation for R360 Global Services (Feb–Mar 2025)
Location: Onsite, Bengaluru
Engaged directly with Excel junkies (financial operations) to gather requirements, design and support development of a financial reconciliation engine for client R360 Global Services.

Large Language Models (LLM)AgentsPython (Programming Language)Systems DesignGenerative AILLM evaluation

Self-employed

AI Engineering

Jun 2024 – Aug 2024 · 2 mos · Remote

Developed a fast RAG solution utilizing Indian government auditor-general data for a stealth startup under tight deadlines; They were able to raise funding later on using this project
Open Source Contribution - CodeQA and educative blog posts; Active in India’s generative AI community

Retrieval-Augmented Generation (RAG)Large Language Models (LLM)

Zuora

2 roles

Software Engineer

Jul 2022 – Feb 2024 · 1 yr 7 mos

◦ Developed and maintained three production codebases supporting multi-million ARR customers as part of Tax team: Tax App (Ruby on Rails microservice), Core Repo (Java), and New Tax App (migrated to Java)
◦ Implemented API integration (Java) for three existing vendors for New Tax App; Designed and implemented from scratch for a new vendor
◦ Led offline validation for three tax vendors for API migration supporting several 1M+ ARR customers
◦ Designed and implemented complex multiple tax percentage rate levels algorithm based on taxation law
changes for a Japanese customer in core codebase
◦ Handled on-call duties, authored oncall runbook, resolved customer issues, and collaborated with team and stakeholders
◦ Reduced logging costs by 60% for old tax app by compressing and concurrently sending logs
Technologies: Java, Spring Boot, Ruby, Ruby on Rails, SQL, Redis, PostGresSQL, Python, React, Docker, Kubernetes;
Tools: Kibana, Grafana, Zendesk, Confluence, Jira, Jenkins

RubyRuby on RailsJavaSpring BootRedis

Software Engineer

Jul 2022 – Feb 2024 · 1 yr 7 mos

Juniper networks

Software Engineer Intern

Jan 2022 – Jun 2022 · 5 mos · Bangalore Urban, Karnataka, India

Developed PoCs in Python to enhance a legacy router output parsing app creating a more generic efficient version
Implemented a parser to extract interface information from configuration files and store in a custom
internal data model and created a compact templating language to enable users to perform conformance checking on parsed data
Learnt about compilers and parsers in the process
Learnt about clouds and containers

Python (Programming Language)C++

Manipal institute of technology

Research Intern

Apr 2021 – Jun 2021 · 2 mos · India

Worked under the guidance of Prof. Swati B.P, ICT dept.
Implemented a hierarchical word and character level model to predict semantic
relationship between a (question,relation,subject) triple from a knowledge graph
Used pre-trained GloVe embeddings and a GRU for word level model
Developed an optimized custom negative sampling procedure for 100x faster training
Implemented Candidate generation process for ranking entities using cosine similarity based TF-IDF vectorizer and Edit distance
Achieved an accuracy of 78% on 5621 test samples of WikiDataSimpleQuestions
dataset
Used Pytorch, torchtext, PolyFuzz, spaCy