Skip to content
←adrij_shikharWork Experience
Agent-native connector development
- Built the surface that lets LLMs generate and operate connectors: a construct library, a connector generator, an installable CLI, a skill library, and an MCP server exposing platform APIs to agents.
- Started the internal agent-tooling platform the team now builds on — repo structure, skill conventions, CI auto-release, marketplace distribution. Other engineers extend it rather than fork it.
Streamlining connector development
- Made connector authentication declarative. A provider used to be its own processor inside the platform — a code change, a database migration, and coordination across three services. It is now a config file, with no platform change at all.
- Moved runtime configuration and config templates into the connector, behind a compatibility kit that gates correctness at build time instead of in review.
- Standardised how connector repos are created, versioned and released, behind a central BOM carrying a named release policy. Before this every connector carried its own build config and drifted.
- Decoupled the connector fleet from the shared platform SDK, so a connector no longer inherits a runtime it does not control.
Connection reliability
- Owned the test-connection contract end to end and took its false-negative rate from over 30% to nil in validation — a failing check now means an actual connection problem rather than a flaky one.
- Delivered configuration validation across every source and destination — Postgres, MySQL, Oracle, SQL Server, Snowflake, BigQuery, Redshift — so a misconfiguration surfaces at connect time rather than at first sync.
- Brought p95 test-connection latency under 10 seconds, and shipped an end-to-end suite alongside each framework rather than leaving the contract unit-tested.
Failure visibility
- Replaced static error classification with a dynamic system. A classification change used to wait on a connector's full release cycle; it now takes minutes instead of 3–7 days.
- Root-caused a class of production hangs to the regex engine rather than the queries being run, and migrated classification to RE2J — which bounds worst-case evaluation to under five seconds and took a load run from a multi-minute stall to 1m19s.
- Built metrics and dashboards for connection health, and put HTTP client metrics into the connector construct library so teams see a rate-limiting API before it becomes a sync failure.
Data correctness at scale
- Delivered SCD Type 2 history mode across Snowflake, BigQuery and Redshift, handling the per-destination SQL differences — Redshift has no
MERGE and needs a four-statement path.
- Built parent-child object handling with inferred deletes, giving full-load objects the delete support they previously lacked.
- Implemented real-time CDC on the Debezium engine, and built the schema catalog service behind it from scratch — schema versioning, metadata management and compatibility checks, so a pipeline survives schema evolution.
- Scaled object handling from 1,000 to 25,000 objects, with a 5x gain in MySQL ingestion throughput.
- Moved every source and destination onto negotiated TLS 1.2 and 1.3, and proved it safe across providers with a dedicated SSL test suite.
Engineering velocity
- Cut build and CI runtime with measured before and after:
mvn clean verify 31:43 → 10:33 and CI 43:32 → 21:24 on the first service, then 11:07 → 5:15 and 5:14 → 2:01 as the approach was adopted elsewhere.
- Ran controlled configuration sweeps against a 90-day baseline rather than tuning by feel, and published the before/after pipelines including the configurations that regressed.
- Built the unified CLI and local development platform — one command, three modes, profiling, a metrics pipeline — so the team runs the stack locally instead of queueing for shared environments.
Earlier platform work
- Implemented near-real-time PII redaction for data governance, on S3 and AWS Comprehend.
- Integrated a microservices architecture on AWS Fargate and a Temporal-based task execution system for hierarchical DAG processing.
- Shipped session logs, YAML pipeline templates, a Cron-based scheduler and Terraform support, and launched the SurveyMonkey connector.
- Unified how HTTP and gRPC failures are reported across the control plane, cutting support overhead.
- Added support for 'wal2json' output plugin for postgres replication.
- Reduced WAL processing time taken by the platform, by 60%.
- Implemented dynamic plugin selection using Guice dependency injection.
- Implemented OAuth 2.0 feature for accessing protected Rest APIs
- Tech Stack: Java, Postgres
- Worked on curating ETL data pipeline from concept to proof of concept.
- Implemented on-demand data transformations using Apache Spark, and streaming the same using Apache Kafka onto Google Cloud Platform.
- Containerized individual components of the pipeline for better development and deployment.
- Solely configured and maintained pipeline on Google K8s Engine.
- Orchestrated data visualization service to configure overlayed charts.
- Tech Stack: Python, Apache Spark, Docker, Kubernetes, Google Cloud Platform
- Collaborated with the core founding team on the initial stages of the platform.
- Setup infrastructure for the applications, keeping scalability and security into account.
- Developed management dashboard service for micro and macro level user access across the apps.
- Customized D3 for visualization of data, fitting our use case.
- Tech Stack: React, Django, Postgres
- Ensured stability of the product by integrating tests and error handling.
- Optimized uploading and validating data from user's end.
- Implemented Stripe, Sentry, Clickup for better development cycle.
- Integrated continuous integration and ensured continuous delivery among various services.
- Worked on graphene to optimize API performance.
- Tech Stack: React, Django, GraphQL, ffmpeg
- Implemented core features and structure from concept through deployment.
- Introduced REST API's, server-side pagination and JWT based authentication system.
- Standardized UI libraries by enclosing them in highly customizable wrapper for code reusability.
- Assessed UX and UI designs for technical feasibility.
- Developed standard and ad hoc report in table format.
- Collaborated with product team members to implement new feature developments.
- Tech Stack: React, NodeJS, MongoDB
- Under the hood of the group, we promote technical culture on the campus by conducting hackathons, lecture series, and competitions.
- Responsible for maintaining current applications and server management.
- Mentored freshmen students in their projects for the Winter of Code program.
- Conducted meetings, hands-on workshops and events on various topics related to entrepreneurship and startups.
- Participated in various case studies regarding SaaS.
- Developed core pipeline and worked on optimizing user experience.
- Spearheaded the development of user interface and the flow of login & registration forms.
- Been a part of a 3-tier team of 5+ executive members, associate members, and co-coordinators to establish web presence of Cognizance 2019 and 2020.
- Contributed as a Manager Web in 2019 and 2020
- Worked on designing the architecture and implementing core features of the progressive web app.
- Ported the legacy code from webpack v2 to webpack v4 and restructured the node dependencies.