Skip to content
View darshansatpathy's full-sized avatar

Block or report darshansatpathy

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
darshansatpathy/README.md

Darshan Kumar Satpathy

AI & Data Solutions Architect in Bhubaneswar, India. I architect enterprise AI that makes it to production: multi-agent platforms, governed data agents, bank risk models, streaming and lakehouse platforms. Portfolio: darshansatpathy.github.io · LinkedIn

Open to Lead and Principal AI Architect roles, including international and remote-first teams.

The repositories below are open-source reference implementations of patterns from my production work. They run on synthetic or public data, are tested in CI and each one measures the failure it is built to prevent.

Agent platform and safety

Repository What it proves
agent-guardrails-gateway Policy enforcement between agents and tools; 23 of 23 attack scenarios contained
multi-agent-ops-automation Supervisor and worker agents behind a deterministic executor with idempotency and approvals
human-in-the-loop-inbox Approvals bound to exact arguments; single-use grants
no-code-agent-studio Agents as YAML, released dev to prod with test gates and risk review
mcp-server-toolkit Production template for MCP servers: OAuth, scopes, tenancy, audit

Grounded and governed AI

Repository What it proves
governed-data-agents Business questions answered without the model writing SQL; row-level security
agentic-rag-reference Permission-aware, cited RAG that abstains when it should
agent-memory-service Agent memory with supersession, PII refusal and tenant isolation

LLMOps

Repository What it proves
agent-eval-harness CI gate that blocks pull requests that make an agent worse
agent-tracing-otel OpenTelemetry tracing built around cost per completed task
llm-router-cost-telemetry OpenAI-compatible router; evaluated on real graded traffic

Data, ML and platform engineering

Repository What it proves
realtime-transaction-streaming Exactly-once Kafka processing, proven by killing processes mid-flight
lakehouse-iceberg-trino Open lakehouse with every table checked against an independent answer
snowflake-migration-toolkit Teradata and Oracle to Snowflake: translation and row-level validation
integration-migration-factory ESB to Kafka migration with record-by-record regression
ha-model-inference-platform Canary releases, SLO burn-rate alerts and zero-drop failover on Kubernetes
db-incident-remediation-agents Prometheus alert to verified PostgreSQL fix, with a policy governor
credit-risk-ml-platform Probability-of-default models with model-risk checks and validation report
demand-forecasting-platform Coherent hierarchical forecasts gated by backtests
lidar-gnn-digital-twin Graph neural network segmentation and change detection on LiDAR scans

Pinned Loading

  1. agent-guardrails-gateway agent-guardrails-gateway Public

    Policy enforcement point between AI agents and their tools: task-scoped grants, argument constraints, human approval, prompt-injection taint tracking, tamper-evident audit log, MCP proxy.

    Python

  2. agentic-rag-reference agentic-rag-reference Public

    Permission-aware agentic RAG: hybrid BM25 + dense retrieval with RRF, cited answers, calibrated abstention, prompt-injection quarantine and an evaluation gate in CI.

    Python

  3. multi-agent-ops-automation multi-agent-ops-automation Public

    Supervisor/worker agents on LangGraph with deterministic tool execution, human approvals, exactly-once side effects and an evaluation gate in CI.

    Python

  4. no-code-agent-studio no-code-agent-studio Public

    Agents written as YAML, statically checked, compiled to LangGraph, and released like software: immutable versions, dev/staging/prod gates, risk review, audit trail, Kubernetes.

    Python

  5. db-incident-remediation-agents db-incident-remediation-agents Public

    Agents that triage and fix PostgreSQL incidents from Prometheus alerts: reviewed runbooks, a governor on every action, approvals, verified fixes and a postmortem per incident.

    Python

  6. governed-data-agents governed-data-agents Public

    Natural-language analytics over a semantic layer: plans not SQL, grants checked before SQL exists, PostgreSQL row-level security, every question audited.

    Python