cat about.txt

About Me

about.txt
cat about.txt

AI Engineer specializing in RL infrastructure and LLM agent systems — built production platforms serving Microsoft and Google, and fine-tuned sub-billion parameter models to within 2% of GPT-4o-mini. Experienced across the full post-training stack: data collection, reward design, agent orchestration, and evaluation on Kubernetes.

cat experience.log

Work Experience

deccanai.log
$DeccanAI
AI & Backend Engineer
Oct 2025 – Present
Hyderabad, India
  • Built a production RL platform (Python, FastAPI) for human-in-the-loop golden trajectory generation, coordinating 500+ annotators and driving $10M ARR.
  • Architected Stark-Forge, a distributed backend framework that runs isolated services on Kubernetes; added pluggable SQL, JSON, and Python verification services with support for multiple agent runtimes.
  • Designed evaluation and verification pipelines by defining success criteria and automated checks for SQL and code tasks, improving end-to-end task success rate from 41% to 68%.
  • Owned delivery of 20+ production MCP servers that generated verified execution traces; the Jira MCP server handles 10,000+ requests/day in production evaluation pipelines used by Microsoft and Google.
  • Fine-tuned Qwen-0.5B for text-to-SQL on Spider using Agent-Lightning, reaching within 2% of GPT-4o-mini accuracy with 3× smaller scale and 10× lower inference cost.
deutsche_bank.log
$Deutsche Bank
Software Engineer
Jul 2024 – Oct 2025
Pune, India
  • Architected a RESTful ETL pipeline (Spring Boot, Airflow, SQL) automating 10,000+ alerts/day across financial workflows; reduced data load time by 40% and freed 3 ops engineers for higher-value work.
  • Engineered high-concurrency microservices with Memcached and Java ExecutorService for business-critical internal systems; reduced outages by 30% and improved P99 latency from 820ms to 310ms for 5,000+ daily users.
aidash.log
$AiDash
Machine Learning Engineer Intern
Jan 2024 – Jul 2024
Bengaluru, India
  • Built a Python ML SDK abstracting Kubeflow and Airflow; compressed pipeline setup from 2 days to 3 hours, adopted by 12+ data scientists with 30% faster experiment cycles.
  • Developed a human-in-the-loop Annotation Service with QGIS for raster validation; eliminated 40% of manual GIS effort across 8 annotation campaigns.
cat education.txt

Education

education.txt
$Indian Institute of Information Technology Allahabad
Bachelor of Technology in Electronics & Communication Engineering
CGPA: 8.80/10.0Nov 2020 – May 2024
cat achievements.json | jq

Achievements

# academic distinction
  • JEE Main 99.34 percentile
  • JEE Advanced AIR ~9,000
# hackathons
  • Top 33 of 1,000+ teams (top 3.3%) at Hack In the North HINT-5.0
  • Rank 2 at Hackout '22
# competitive programming
  • Codeforces 1549 (Specialist, top 15%)
  • CodeChef 1982 (4-Star, top 5%)
  • Google Kick Start 2022 rank 1636
ls skills/ --verbose

Skills Matrix

# AI & ML
Reinforcement LearningLLMsSFTRLHFPost-trainingFine-tuningReward DesignEvaluation
# Agent Systems
LangChainAutoGenCrewAIAgent-LightningMCPTool-use AgentsMulti-Agent Orchestration
# ML Infra & Cloud
KubeflowAirflowDistributed TrainingKubernetesAWSGCPDockerJenkinsCI/CD
# Languages
PythonJavaTypeScriptJavaScriptC/C++SQL
# Backend
FastAPISpring BootDjangoREST APIsMicroservices
# Databases
PostgreSQLMySQLOracleMemcached
# Concepts
Distributed SystemsSystem DesignDSAAsync ProcessingData Pipelines