CV

AI infrastructure & distributed systems · GitHub AI Engineering, Microsoft · Zürich

📧 sameerankalgi@gmail.com · 💼 LinkedIn · 🐙 GitHub

I build the layer that decides whether a model ships or stalls — GPU inference, distributed systems, and the platforms underneath them. 12+ years across the stack: Kubernetes for connected cars, classified-grade cloud for a national army, high-throughput transaction systems, and a healthcare startup I co-founded and ran as CTO.

Currently

Microsoft — GitHub AI Engineering · Senior Software Solution Engineer

Zürich

• Benchmarked AlphaFold 2 on Azure H100 GPU HPC clusters for a pharma customer — profiled the protein-folding pipeline end-to-end (MSA search, Evoformer, structure module) and tuned throughput until large-scale structure prediction was viable on their infrastructure.
• Served DeepSeek across multi-GPU A100 clusters with vLLM and tensor parallelism, tuned for production throughput and latency.
• Advise enterprise customers on designing, deploying, and scaling GPU-based LLM inference and AI workloads on Azure.
• Work across GitHub AI Engineering, bringing Copilot and AI tooling into enterprise developer workflows — including WRK541 at the Microsoft AI Tour Zürich.

Previously

Swisscom — Software Architect / Technical Lead

Zürich

• Architected secure, classified-grade cloud infrastructure for the Swiss Army.
• Led cloud-transformation architecture for enterprise clients, improving scalability and performance across distributed systems.
• Built GenAI automation solutions for manufacturing SMEs.

Peech Care — Co-Founder & CTO

A speech-therapy platform that detected speech disorders — stammering, lisp, and other articulation disorders — from users’ voice notes, and recommended clinically-vetted therapy. Built to cut Germany’s long therapist waiting times and lower the stigma around asking for help, for children and adults alike.

• Co-founded the company and led engineering as CTO — owned hiring, technical architecture, and product direction for a team of 5 engineers.
• Built the core ML pipeline: extracted acoustic and frequency features from voice notes to detect articulation disorders and recommend therapy vetted by certified speech therapists. (Approach informed by research on acoustic stuttering-event detection — e.g. Lea et al., “SEP-28k,” ICASSP 2021.)
• Designed the platform to train models on anonymized user data, with a roadmap toward gamification and avatar-guided therapy.
• Shipped a full MVP — app and backend — and secured €150K in funding.

Daimler TSS (contract, via Devoteam) — Technical Lead

Stuttgart

• Migrated Daimler’s connected-car data platform off an IBM Db2 column-store onto MongoDB running on bare metal — ~2 TB and 600M+ records.
• Built the migration service in Go — distributed by design, with idempotent retries and end-to-end integrity checks — cutting over production data with zero downtime and no data loss in a 33–38h window (~4–5K writes/sec sustained).
• Deployed and hardened the MongoDB cluster on Red Hat Linux for production — disabled Transparent Huge Pages, configured SELinux, and owned sharding across the fleet.
• Deployed the Go microservices on Kubernetes and scaled them end-to-end, owning rollout, autoscaling, and reliability.
• Managed the bare-metal fleet’s full OS lifecycle — kernel upgrades, patching, and system tuning across the cluster.

BCG Platinion — Senior Software Engineer

Berlin

• Built an invoice-processing system for Henkel — Python, PostgreSQL, Bootstrap front end. Designed and built the backend and containerized the microservices.
• Deployed ML pipelines powering a chatbot within the system.
• Sanitized vendor data from the client — wrote the consistency and integrity checks that ran before anything reached PostgreSQL.
• Led development with a cross-functional 5-member team, balancing client needs on functionality, timeline, and performance.

PwC — Senior Consultant

Berlin

• Planned and developed finance/audit applications with Vue.js, Angular, Spring Boot, and Java — augmenting PwC’s Aura audit platform with automated document processing.
• Directed a team of 5 cross-functional consultants, driving continuous improvement of the auditing software across client implementations.
• Managed an offshore team in India as project manager, porting an application from VB to Java.
• Wrote about the pitch-club career-switch moment on Medium.

Intelligent Data Analytics (IDA) — Software Engineer

Frankfurt

Recruited via the PCDE pitch-club community — the jump that took me from consulting back into shipping code.

SAP — SAP HANA Developer

• Built a backend system to process raw event data into custom reports on SAP HANA Cloud Platform using HANA XSJS.
• Worked on the SAP Web Analytics tracker, monitoring framework changes.

BookMyShow — Software Developer

• Built a SQLite-based transaction-queuing system to buffer requests under poor connectivity, improving throughput.
• Replaced an RDBMS login backend with MongoDB, reducing login failures and improving scalability.
• Built a raw-data processing web service (JSON/XML) with ZeroMQ, Node.js, and MongoDB, reducing system latency and load.

Stack

Selected artifacts

• pi-bench — composed-defense benchmark for prompt injection. Grades whole defense stacks on attack success rate, false positives, latency, and cost.
• oss-model-playbook — enterprise playbook for open-weight LLMs on on-prem Kubernetes, AKS, and Azure AI Foundry.
• vector-engineering-for-agents — embeddings, ANN, hybrid retrieval, agent memory, and production ops for RAG.

Publication

• Reactive Programming Languages — A Survey

Research interests

• GPU inference at scale: batching, KV-cache pressure, tensor/pipeline parallelism, the gap between benchmark throughput and served p99
• AI safety as engineering practice: containment, audits, choke points (in the Coming Wave sense)
• Agent eval harnesses: failure modes, trust boundaries, reversibility
• Enterprise AI infra: the “boring layer” that decides whether AI ships

I review papers under /ai-safety — currently working through alignment evals and interpretability literature.

Certifications

KCNA — Kubernetes and Cloud Native Associate (CNCF) · MongoDB DBA · MongoDB & Microsoft Champion · B1 German

→ Kubestronaut progress — 1 of the 5 CNCF Kubernetes certifications, KCSA next.

Speaking

Available for technical talks on:

• GPU inference and open-weight model serving in the enterprise
• AI agents in regulated environments
• The transition from consulting to engineering (and back)

→ Get in touch.

Languages

English · German · Hindi · Kannada

Outside work

Cycling around the Zürichsee, running along the Limmat, occasional travel write-ups.

PDF version on request.