About

About

I’m Ankit Kumar, a Senior Software Engineer who builds low-latency, large-scale streaming and ML infrastructure. My work spans the full stack of a modern CTV/ad-tech platform — from squeezing latency out of hot paths and reducing lock critical sections, to owning the Kubernetes and data-pipeline layers that hundreds of jobs and millions of content items run on.

I care about two things in roughly equal measure: systems that are fast (careful memory, concurrency, and I/O work) and systems that scale and stay up (distributed data pipelines, platform ownership, and ML infrastructure). I hold a B.Tech in Mathematics & Computing from IIT Delhi and have published research in databases and NLP.

📄 Résumé (PDF)   GitHub  ·  LinkedIn

The résumé PDF is a starting point — replace assets/resume.pdf with your latest, Anoki-current version.

Experience

Senior Software Engineer — Anoki AI

Gurugram, India · Sep 2024 – Present

  • ML / vector-search infra: designed and built a PostgreSQL-backed semantic content-search system from scratch — provisioned the RDS instance, schema and indices, migrated embeddings into Postgres, and ran a formal evaluation of query relevance and latency/QPS at 50K → 150K content scale. Deployed embedding servers on EKS/GKE for the “copilot” feature, stood up a GPU-based split server for video search, and built periodic embedding-refresh pipelines.
  • Low-latency & systems performance: reduced lock critical-section time by reading MaxMind geo data straight from bytes; cut cache memory footprint and added multi-level DB caching; load-tested the config server and rate-limited HAProxy; chased down memory growth, intermittent connection drops, a readBlocking segfault, and native crashes on the C++/Android ACR client.
  • Platform / infra ownership: own the Kubernetes deployment layer for the whole platform (ArgoCD + Kustomize, consolidated manifests, resource tuning) and the Airflow orchestration (prod upgrades, Consul → Airflow-variables migration, OpsGenie on-call alerting). Ran a disciplined phased production rollout (1% → 10% → 50% → 100%) with monitoring, executed a Graviton/ARM64 migration of recorder/config-server/Airflow pods (incl. custom ARM64 Fluentd images) for cost and performance, and authored the org-wide Kafka client libraries in Go and Python adopted across services.
  • Streaming systems at scale: built the EPG schedule-recorder + stitcher (decoupled scheduler/recorder, HLS export, MP4 generation, SCTE ad-marker detection, Kafka metadata ingestion, retry/backfill) and a live HLS playlist server (snapshot APIs, segment stitching from Kafka, HTTP → Kafka ML interface, CloudFront delivery, safe/unsafe brand-safety split).
  • Ad-tech data & fraud/brand-safety: built Invalid-Traffic (IVT) detection end-to-end — global and advertiser-level aggregation, Superset dashboards, periodic Airflow blocklist/stats jobs, and blocklist push to Beeswax. Spark/Scala aggregation for viewer insights, content intelligence, app-viewership and geo-distribution pipelines.

Senior Software Engineer → Software Engineer — LG Ad Solutions (formerly Alphonso)

Bengaluru, India · Feb 2021 – Aug 2024 (SWE, then Senior SWE from Apr 2023)

  • Data Services (Spark, Scala, Python, Go, HDFS, Kafka, Elasticsearch, Airflow): set up HA Airflow infra managing 100s of jobs; owned Elasticsearch infra scaling to 100s of TB; built an EPG pipeline processing 100s of GB from multi-country providers; owned a Spark-Streaming ETL from Kafka → Elasticsearch/HDFS at 500K records/min.
  • Server-Side Ad-Insertion (Go, Node.js, Redis, Kubernetes, HLS, MPEG-DASH): built an ad-insertion server from scratch that scaled to 400K+ concurrent live viewers (profiled with pprof/Clinic.js/K6); JIT VoD ad-insertion improved ad watch-completion by 50%; custom internal interfaces improved end-to-end performance by 75%; multi-level caching added 50% headroom under load.
  • Content Hosting & Management (Go, Node.js, MySQL, Elasticsearch, Kafka, Kubernetes, React): delivered 6,000+ VoD and live content pieces; built a multi-partner onboarding lifecycle and an API + UI serving over 1M+ content items.
  • Smart-TV Operating System (C++, Node.js, Android): built a resilient downloader/updater (100s of MB over weak links, surviving reboots), a socket-IPC app-manager, and a 100% automated test-suite across Linux and Android TVOS.
  • Rated a top performer; mentored an intern into a full-time role; co-organized Code Like Ada, a women-in-tech hackathon.

Software Engineering Research Intern — University College London

London, UK · May – Jul 2020

  • Used Snorkel to combine code-smell detectors (PMD, Infer, Error-Prone) into a strong labeller and proposed a self-attention Transformer-based bug-detection framework.

Database Research Intern — Imperial College London

London, UK · Dec 2019 – Apr 2020

  • Proposed hybrid learned spatial indexes (Interpolation-friendly R-Tree, KD/Quad/Oc-Tree), reducing query time by 60% while shrinking the index memory footprint by over 90% — published as Hands-off Model Integration in Spatial Index Structures (AIDB@VLDB, 2020).

Software Engineer Intern — Elucidata

New Delhi, India · May – Jul 2019

  • Built an Abstract-Factory-based package generating Jupyter notebooks from YAML, cutting data-analysis time by 95%.

Education

Indian Institute of Technology, Delhi — B.Tech, Mathematics & Computing (major), Computer Science (minor) · 2017 – 2021 · CGPA 9.0/10

Skills

  • Languages: C/C++, Go, Scala, JavaScript/TypeScript, Python
  • Data & messaging: PostgreSQL, MySQL, Elasticsearch, Kafka, Redis, MongoDB
  • Frameworks & data: Spark, React, Node.js
  • Build & deploy: Docker, Kubernetes, ArgoCD, Kustomize, Jenkins, Istio, Helm
  • Observability: Prometheus, Grafana, InfluxDB, Chronograf, Kibana

Publications

  • Identification of Misinformation in COVID-19 Tweets Using BERTweet — NLP4IF, 2021
  • Hands-off Model Integration in Spatial Index Structures — AIDB@VLDB, 2020

Honors & Awards

  • JEE 2017 — All India Rank 226
  • CBSE Mathematical Olympiad 2015–16 — All India Rank 11 (top 35 nationally)
  • SOF International Mathematics Olympiad — International rank < 25, three times (2014–16)
  • KVPY Scholar (top 1%) · NTSE Scholar (top 1000 nationally)
  • Teaching Assistant (Database Systems, Intro to CS) · Academic Mentor for 100+ first-years
  • Technical Coordinator, Rendezvous 2019 (built the festival’s main app)