Shreyash Pandey // Software Development Engineer 2 at IBM
I build AI systems, automation workflows, and backend products that survive production
My recent work sits where applied AI meets engineering discipline: locator auto-healing, semantic retrieval, internal copilots, multilingual model training, and automation pipelines with real operational guardrails.
- 5K+ test locators auto-healed
- 3-tier ML plus VLM recovery design
- Hackathon-winning semantic search
- Open-source model training from scratch
Current proof surface
The strongest signal in the portfolio is not a single project. It is the combination of AI recovery design, service reliability, and hands-on model building shipped across work and independent systems.
What I optimize for
The common thread across my work is reliability under ambiguity. I like systems that need to reason, recover, and still stay operationally legible.
Start with the cheapest reliable fallback
I design systems to attempt deterministic recovery first, then graduate into ML and model-based fallbacks only when they are justified.
Treat evaluation as part of the product
When the system contains AI, the measurement loop is not optional. I care about observable accuracy, drift, error budgets, and failure analysis.
Systems I have shipped
2024 - Present
Software Development Engineer 2
IBM Software Labs · Bengaluru, India
I design and ship reliability-heavy AI capabilities inside browser automation and testing products. The work combines embeddings, vision-language models, service decomposition, and a lot of operational discipline.
2023 - 2024
Software Engineer 2
Software AG (now IBM) · Bengaluru, India
This phase pushed me deeper into AI product work: semantic retrieval, internal copilots, and prediction systems grounded in practical product needs rather than demos.
2022 - 2023
Software Engineer
Software AG · Bengaluru, India
I worked on enterprise integration platform capabilities across Java, Spring Boot, and REST APIs, building the foundation that still shapes how I reason about production systems.
Current projects
2026 · Decoder-only language model
Phoenix 125M
A LLaMA-style 125M parameter model trained from scratch on a single RTX 3080 Ti with a custom tokenizer, data pipeline, and training loop.
2026 · Multilingual language models
Sweta-Hi and Sweta-Kn
Hindi and Kannada pretraining efforts built on a LLaMA-style architecture with custom tokenizers and an end-to-end multilingual data pipeline.
2026 · Fine-tuning · Text-to-SQL
SQLForge: Mistral 7B QLoRA
A 4-bit QLoRA fine-tune that turns Mistral 7B v0.3 into a reliable text-to-SQL model. The same 12 GB GPU used for Phoenix 125M, with a 3.75 GB VRAM headroom budgeted up front, and a schema-aware evaluation rebuild after the first WikiSQL run showed the metric was lying.
See all builds
Explore the full project portfolio
Looking for my next role
AI Systems Engineer based in Bengaluru, India
I am most interested in roles where AI systems, backend engineering, and reliability work intersect. That usually means agent infrastructure, evaluation-heavy product work, automation platforms, or developer tooling.