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.

Enterprise architecture, cloud and governance

Repository What it proves
azure-ai-landing-zone Private, keyless Azure OpenAI in Terraform, with a policy-as-code checker that fails the build if a door is reopened
ai-platform-reference-architecture A 27-component AI platform written as data; archfit fails the build when the design contradicts itself
ai-governance-evidence-pack 41 EU AI Act, NIST AI RMF and ISO 42001 controls mapped to hash-pinned evidence, with gaps stated openly
databricks-medallion-lakehouse Each table is a spec; Lakeflow SQL, Unity Catalog governance and a bundle are generated and linted (not yet run on Databricks)
claude-agent-reference A production agent loop on the Claude API in about 2,000 readable lines, zero runtime dependencies

Delivery governance for AI systems

Repository What it proves
ai-design-review-kit An AI architecture review as files: questions, evidence, accepted risks and a verdict CI can enforce
ai-threat-model-as-code Threats derived from the data-flow graph and mapped to OWASP LLM Top 10 and MITRE ATLAS ids; ready, conditional or blocked
model-promotion-gates 23 evidence-bound gates on model stage transitions, with approvals tied to the exact artefact digest and expiring waivers
data-contracts-registry Data contracts as reviewed code; a change that would break a consumer fails the build
ai-finops-showback AI spend by team and app from usage exports, shared costs allocated by explicit rule, totals that agree to the cent

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. ai-governance-evidence-pack ai-governance-evidence-pack Public

    EU AI Act, NIST AI RMF and ISO/IEC 42001 controls mapped to hash-pinned evidence in real repositories, with honest gaps.

    Python

  4. ai-platform-reference-architecture ai-platform-reference-architecture Public

    Governed enterprise AI platform kept as data and checked by architecture fitness functions; risk register and operating model.

    Python

  5. azure-ai-landing-zone azure-ai-landing-zone Public

    Private, keyless Azure foundation for AI workloads with a Terraform policy checker (tfpolicy).

    Python

  6. model-promotion-gates model-promotion-gates Public

    Evidence-bound promotion gates for ML model stage transitions: approvals tied to the artefact digest, expiring waivers, allowed, conditional or denied verdicts.

    Python