Jan 2025 – Apr 2025 · 3 mos · San Francisco Bay Area · Remote

• Seed stage ($9.5M seed round) startup with an exceptional team of people from Stanford, EPFL, Google Research, OpenAI and more
• Owned reliability, and performance for a key product with >200k users. This involved one full rewrite of deployment infra, lots of firefighting, and late night debugging :)
• Rewrote the container-orchestration, deployment infra, and improved observability, reducing downtime by ~5x
• Optimized the customer-facing vector search, and text search latency by 2x, by improving caching, and aggregation logic
• Implemented a decoupled token-based auth service, resolved concurrency bugs, performance issues and security vulnerabilities

Jul 2024 – Aug 2025 · 1 yr 1 mo · Germany · Remote

• Making WebAssembly (wasmtime serve) scale over 3 times better on modern hardware (10s of cores, multiple NUMA nodes) — nominated for best applied thesis
• Played around with perf, kernel source (memory subsystem), profilers, and wrote really fast C++, with per-core sharding with Seastar
• Profiled, and deduced the bottleneck to be TLB shootdowns causing Inter Processor Interrupts (IPIs)
• Implemented a shard-per-core serving infrastructure in C++, with provision for read-only shared memory (think heavy neural network weights)
• Achieved 3-4x increase in throughput, without compromising fast instantiation by retaining Copy-on-Write and guard-pages

May 2024 – Jul 2024 · 2 mos · New Delhi, Delhi, India · On-site

• APT Portfolio is a leading High Frequency Trading (HFT) firm, operating in Indian markets since 2009, and in Europe and America since 2015
• So far the most learning (and fun) I’ve had in any internship. Learnt about components of a trading system, Clickhouse internals, orderbook implementation and nanosecond latency logging :)
• Developed a framework to analyze effects of factors like OS, processor, and build flags on latency of all components of trading system
• Reported 2 critical issues, and more than 5 performance regressions in live trading system, using the framework & custom statistical test
• Proof-of-concept tested post-link optimisation, achieving upto 10% latency reduction, leading to production trials by senior engineers

Aug 2023 – May 2024 · 9 mos · On-site

• Optimizing context switches in IO intensive environments (think Redis, Nginx, and Backend containers). In collaboration with IRDBT.
• Patched, profiled, (ab)used, and played around with Linux IO subsystem, and the syscall interface :)
• Worked on runtime syscall optimization, for Linux kernel, using dynamic binary-to-binary translation from user-space to kernel-space
• Developed a loadable kernel module, to intercept syscalls, and performed ablation studies on previous userspace-bypass research
• Demonstrated throughput increase of 250% on IO tight-loops, and 30% on Redis, while observing performance degradation in containers

Jun 2023 – Dec 2023 · 6 mos · San Jose, California, United States · Remote

• MVP stage startup founded by industry-leading KGP Alumni from 80's and 90's in San Jose, California
• Part of Trumio Pioneers, the initial hires to build, and validate the platform. Worked directly with the amazing co-founders :)
• Developed a chat-based co-pilot, for requirement generation and querying for an MVP stage startup
• Backend with FastAPI, OpenAI LLMs, and FAISS vector store, served through Azure
• Built out a search and recommendation system using vector search over MongoDB

Oct 2022 – Jan 2023 · 3 mos · On-site

• SEAL is the hardware security laboratory @ IIT Kharagpur, headed by Prof. Debdeep (Bhatnagar prize recipient)
• Worked on FPGA Programming in Verilog to modify and “lock” an open-source RISC Processor.
• Metaprogramming and code generation-based approaches to generate locking code.

Jun 2022 – Nov 2022 · 5 mos · Kharagpur, West Bengal, India

• KDAG is the student-led data science research group of IIT Kharagpur.
• Overseeing the execution of all activities of the Kharagpur Data Analytics Group. And coming up with new initiatives for the KGP Data Science Community.
• Conducted a Neural Art Workshop, with 800 registrants, and 7 speakers, from 3 countries, and domains across architecture, art generation, and VR!

Aug 2021 – Jun 2022 · 10 mos · Kharagpur, West Bengal, India

• Promoting and imparting knowledge of Data Analytics to the KGP Community.
• Part of the organizing team of the RedBus Datathon.
• Taught in the Machine Learning Fundamentals workshop.
• Wrote Medium articles on Data Science fundamentals.

Apr 2022 – Sep 2022 · 5 mos · Wilmington, Delaware, United States · Remote

• AdPushup is one of the fastest-growing startups in the US (by Financial Times), and winner of the NASSCOM50 award.
• Optimized the ad-bidding pipeline serving 60M data points daily, by predicting bidder participation given minimal user data.
• Iterated over 15+ classification pipelines like SVM, and Factorization Machine in Apache Spark, achieving 84% accuracy.
• Read 30+ research papers, and built a proof-of-concept using a Collaborative Filtering Recommendation Engine.
• Optimized the collaborative filtering via Alternating Least Squares, and achieved an AU-ROC of 0.96 and AU-PR of 0.87.

Nov 2021 – Jan 2022 · 2 mos · Wilmington, Delaware, United States · Remote

• AdPushup is one of the fastest-growing startups in the US (by Financial Times), and winner of the NASSCOM50 award.
• Benchmarked, and setup the Kafka cluster to migrate a pipeline, improving its throughput from 60k, to 220k (messages/sec).
• Developed and fine-tuned a concurrent Kafka Producer in Python, to use for benchmarking, and optimizing the cluster.
• Administrated the Azure Resource Group for testing. Wrote a Bash test suite running 20 producer instances per machine.
• Optimized to achieve a throughput of 220k messages/sec on a 3 node Kafka cluster, and 95k messages/sec on a single node

Jun 2021 – Aug 2022 · 1 yr 2 mos · Kharagpur, West Bengal, India

• Polished the fundamentals of Finance, and Data Analytics.
• Presented a white paper on "Chatbots in Trading", covering scalable chat and bid frameworks.
• Working on research paper implementation, and development of trading algorithms.
• Worked on the development of a custom backtesting software, with the Senior Quant Researchers.