Software Engineer @ Microsoft — building applied-LLM products (multi-agent orchestration, RAG, LLM evaluation) | backend & distributed systems | ex-Goldman Sachs

Bengaluru

Bengaluru

Remote
Bengaluru

Bengaluru, Karnataka, India



Integrated Dual Degree, Computer Science
Indian Institute of Technology (Banaras Hindu University), Varanasi
Join Shubham and 10k+ other professionals on
Tal is the app where bosses meet engineers & builders directly, without middlemen
Download Tal BossSoftware Engineer @ Microsoft — building applied-LLM products (multi-agent orchestration, RAG, LLM evaluation) | backend & distributed systems | ex-Goldman Sachs
I build applied-LLM systems and the backend platforms underneath them. Right now that means working on an enterprise HR Copilot at Microsoft, where I've built a multi-agent orchestration MVP and subsequently into the platform, added retrieval-augmented generation (RAG) for organizational-data questions, and automated LLM evaluation so the team can iterate in minutes instead of days. The through-line across ~9 years is owning systems end-to-end — from the architecture argument through to the thing running in production — and caring as much about what not to build as what to build. Some of the work I'm most satisfied with is the work that made a large build unnecessary, or turned a scrappy prototype into something durable and reusable. I started out in finance at Goldman Sachs, spent time as a founding engineer at an early-stage startup (Auxia) owning data-pipeline and event-schema work at 4 TB/day scale, and have led small engineering teams along the way. My foundation is data structures, algorithms, and distributed systems; the applied-AI work sits on top of that. I don't think LLMs reason perfectly — they reason fairly well, and most of the interesting engineering is in the scaffolding around them: routing, grounding, evaluation, capacity. That's the space I like working in. Currently based in Bengaluru. Happy to talk applied LLMs, agent architectures, or backend systems at scale.
Greater Bengaluru Area
claude code
github copilot
Software Engineer 2 at Microsoft, May 2024 - Present · 2 yrs 5 mos, Bengaluru, I work on an enterprise HR Copilot at Microsoft. I built retrieval-augmented generation (RAG) into the Copilot so it answers organizational-data questions grounded in each tenant's own data instead of the base model's static knowledge. I also automated LLM evaluation to take iteration from days to minutes, and spent six months owning capacity planning for the service. I built a multi-agent orchestration MVP that showed clear latency and answer-quality gains, and subsequently into the platform.
Software Engineer at Auxia, Sep 2022 - Feb 2024 · 1 yr 6 mos, Bengaluru, I was the first engineer at an early-stage startup and led a small team of three. When a large new customer pushed a 4 TB/day pipeline past its hard 4-hour SLA, I brought it back within bounds by moving the group-by out of map-reduce into upstream SQL. I also owned an event-schema revamp end to end — redesigning the database schema to carry structured event attributes, implementing it, and building the customer-facing UX for choosing which attribute combinations to aggregate. As the team's generalist engineer, I worked across our gRPC/protobuf microservices on Kubernetes.
Software Engineer at Programming Pathshala, Jan 2022 - Aug 2022 · 8 mos, Remote, I designed and led a three-engineer team to ship a real-time WhatsApp notification system on Kafka, which cut first-response time on student doubts from about ten minutes to three. I also did database tuning, introducing a session pool to fix connection starvation and resolving Postgres transaction contention.
Founder, software engineer at Qmind, Jun 2019 - Nov 2021 · 2 yrs 6 mos, Bengaluru, I founded and built a personalized quiz platform for high-school students: a Flutter web app on GCP with Firestore, per-student topic-wise performance tracking, and a rule-based personalization engine. To keep the question bank current I automated question extraction from source PDFs with PyMuPDF. I taught alongside building it during this period.
Analyst at Goldman Sachs, Jul 2017 - Jun 2019 · 2 yrs, Bengaluru, Karnataka, India, I parallelized a multi-hour, PR-gating test suite down to under three minutes, unblocking the team's merge flow. I also built market-risk models on Goldman's proprietary pricing infrastructure and a Kafka-based distributed job that computed price corrections for a class of market instruments.
Integrated Dual Degree, Computer Science, Indian Institute of Technology (Banaras Hindu University), Varanasi