Software Engineer.AI systems, data pipelines,and the harnesses that keep them honest.
MS Information Systems, Northeastern University. Based in Boston. I build event-driven multi-agent systems, warehouse pipelines, and the evaluation harnesses that make non-deterministic systems measurable.
▸ What you are looking at is Argus, one of my systems. A pull request fans out to four specialist agents on independent queues, then aggregates into a single review. Hover a node, or click anywhere.
I work across AI systems, data engineering, and backend infrastructure. Most of what I build sits at the intersection: event-driven multi-agent architectures, retrieval over messy real-world documents, and warehouse pipelines with the quality checks that keep downstream numbers trustworthy.
I care about the unglamorous parts, evaluation, observability, failure isolation, because those are what separate a system that demos well from one that survives contact with production. I use AI coding tools daily and have opinions about where they help and where they get in the way.
- Currently
- MS Information Systems, Northeastern. May 2026
- Based in
- Boston, MA · Open to relocation
- Focus
- Agentic AI · data engineering · backend infrastructure
- Daily tools
- Claude Code · Codex · Cursor
- Looking for
- New grad & mid-level software engineering roles
Every number on this site is one I can point at in a repository
- 4.0/4.0
- GPA, MS Information SystemsNortheastern University
- 20
- public repositories shippedevery project on this site links to its source
- 2
- peer-reviewed publicationsnetwork threat detection, both with DOIs
- 26
- engineers taught as TAacross 9 concurrent team projects
Fourteen systems, each with its architecture and the decisions behind it
Every case study walks the real topology (drawn from the code, not the README), and every number links back to something checkable in the repository. The full index has all twenty.
Graduate Teaching Assistant. DAMG 7374
Northeastern University
Jan 2026 – Apr 2026 · Boston, MA
- Teaching assistant for Data Engineering: Impact of Generative AI with LLMs, under Professor Kishore Aaradhya, supporting 26 engineers across 9 concurrent team projects.
- Administered the Snowflake environment underpinning all nine projects, provisioning role-based access and compute warehouses, and resolving reliability issues before they blocked delivery.
Machine Learning Research Assistant
CVR Research Lab
Jan 2024 – May 2024 · Hyderabad, India
- Developed and evaluated classification systems in Python and PyTorch for network threat detection, through iterative feature engineering and model tuning.
- Published two peer-reviewed papers documenting system design, tradeoffs and evaluation methodology, one on DDoS detection using SMOTE-balanced classifiers, one on Android malware detection with deep neural networks.
Full-Stack Engineering Intern
AIThinkers
Jan 2023 – Dec 2023 · Hyderabad, India
- Shipped production full-stack features and backend APIs at an early-stage startup, owning delivery from specification through deployment and monitoring.
- Diagnosed a silent data-corruption defect through systematic log analysis and profiling, then shipped a targeted fix that eliminated a recurring class of downstream errors.
Five things I believe about building systems that hold up
- 01
Fix the process before automating it
An automated bad process is a faster bad process. On MyClaw I redesigned the manual workflow for correctness first, then built the agentic layer on top of something that already worked.
- 02
Evaluation is infrastructure, not an afterthought
Non-deterministic systems need harnesses the way deterministic ones need tests. That means curated cases, a scoring rubric, and a scheduled job that catches regressions, plus validating the judge itself against human labels before trusting its numbers.
- 03
Failure isolation over monolithic design
Argus gives every specialist agent its own queue and its own dead-letter queue, so one slow reviewer cannot stall a review. Independent failure domains beat a single prompt that does everything.
- 04
If you cannot see it, you cannot operate it
Trace context that survives a Kafka boundary, structured logs carrying trace IDs, and alerts on the signal that matters: one user failing five times is not the same incident as five users failing once.
- 05
Claim only what you can show
Every number on this site is one I can point at in a repo. Where I have not measured something yet, I say so instead of estimating.
What I actually use
Everything here appears in at least one linked repository. Ordered by how central it is to the work on this site, not alphabetically.
- Languages
- PythonTypeScriptJavaScriptSQLGoJavaSwiftC#R
- AI & LLM Systems
- LangGraphLangChainDSPyRAG pipelinesMCP serversagent orchestrationtool & function callingLLM-as-judge evaluationjudge validationchunking & embeddingsprompt engineeringClaude APIOpenAI APIAmazon BedrockVertex AISnowflake CortexPinecone
- Data Engineering
- AirflowdbtSnowflakeBigQueryKafka / RedpandaSpark & PySparkDatabricksETL & ELTdimensional modellingmedallion architecturedata lineageGreat Expectationsdata quality
- Backend & Full-Stack
- FastAPINode.jsExpressSpring BootFlaskDjangoRESTgRPCSSE streamingReactNext.jsPostgreSQLMySQLMongoDBRedismicroservices
- Cloud & DevOps
- AWS (Lambda, SNS, SQS, DynamoDB, ECS, EC2, RDS, S3, Bedrock)GCP (Cloud Run, Cloud Functions, Pub/Sub, Firestore, BigQuery, Vertex AI, GKE)Azure Blob StorageTerraformAWS CDKDockerKubernetesGitHub ActionsCloud BuildGit
- Observability & Quality
- OpenTelemetrySigNozClickHousedistributed tracingtrace context propagationstructured loggingmetrics & alertingSLOsunit & integration testingcode reviewroot-cause analysis
- Security & Governance
- RBACleast-privilege IAMPII & PHI masking (Presidio)input/output guardrailscitation faithfulness checksaudit loggingapproval-gated automationsecure codingGDPR & CCPA awareness
- ML
- PyTorchHugging Facescikit-learnPandasNumPyPEFT & LoRA fine-tuningfeature engineeringmodel evaluationCohen's kappaNLP
- BI & Analytics
- Power BITableausemantic layersself-serve analyticsmetric definition governance
- AI Development Tools
- Claude CodeCodexCursor
Degrees
Northeastern University
4.0 / 4.0Master of Science, Information Systems
Aug 2024 – May 2026 · Boston, MA
CVR College of Engineering
9.2 / 10.0Bachelor of Technology, Computer Science
Data Science specialisation · Cybersecurity minor
Aug 2020 – May 2024 · Hyderabad, India
Peer-reviewed publications
- 2023
The Investigative Study on the Performance Analysis of SMOTE employed Machine Learning Classifier Models to DDoS Attack Detection
International Journal on Recent and Innovation Trends in Computing and Communication, 11(9), pp. 702–708
DOI 10.17762/ijritcc.v11i9.8862 ↗
- 2025
Malware Detection in Android Applications Using Deep Learning Neural Networks
Book chapter, CRC Press / Taylor & Francis
DOI 10.1201/9781003616252-41 ↗
What I'm building this month
- Building
- DocIntel, serverless document intelligence on GCP, built level by level from a single container up to least-privilege service accounts, Terraform and CI. Most recently rewrote the processor from Python to Go on the same trigger. Source ↗
- Thinking about
- How to make evaluation harnesses cheap enough that they get built before the system rather than after it. The Lincoln judge-validation work convinced me most LLM-as-judge numbers are unexamined. Source ↗
- Next
- Adding a timing harness to Argus so its end-to-end review latency becomes a measured number instead of one I decline to publish.
- Looking for
- New grad and mid-level software engineering roles in AI systems, data engineering or backend infrastructure. Boston-based, open to relocation.