Senior Software Engineer · Hyderabad, India

I engineer the systems around the model.

I build production agent runtimes across orchestration, context, tools, trace-driven evaluation, and security controls—turning foundation models into reliable software for complex enterprise workflows.

View selected work View résumé
  • Agent runtime
  • Context engineering
  • Harness engineering
  • Reliability & security
production_agent / runtime trace system active
TRACE / 0248REGULATED WORKFLOW
PII MASKEDPOLICY ACTIVE

ACTIVE DECISION / 01

Curate the working state

Load only the instructions, evidence, session state, and domain constraints required for the current decision.
Langfuse traceEvaluation suiteGuardrail gate
CONTEXT → DECISION → ACTION → GUARD → EVALUATE → IMPROVE

01 / SELECTED WORK

Systems built for everything after the model call.

Anonymized case studies focused on architecture, engineering decisions, reliability, and system-level outcomes.

01

Agent infrastructure

Production Agent Runtime

A modular runtime for evidence-heavy, regulated workflows—designed around orchestration, context assembly, skills, validation, and response generation rather than a single prompt chain.

  • Designed specialized flows across orchestrators, skills, validators, context, and response layers.
  • Separated shared runtime behavior from workflow-specific capabilities so the system can evolve without prompt-level coupling.
  • Built controlled transitions between reasoning, tool use, validation, and final response generation.

System outcome

Created an extensible foundation where new agent capabilities can be added without rebuilding the core loop.

  • Agent runtime
  • Orchestration
  • Skills
  • Validation
02

Evaluation, observability, and safety

Agent Reliability & Security Layer

A layered reliability system that makes agent behavior observable, regressions testable, and unsafe inputs detectable before they reach critical workflows.

  • Integrated Langfuse traces to inspect complete agent trajectories, tool calls, latency, and failure paths.
  • Built an internal evaluation suite for scenario-based quality checks and repeatable regression testing.
  • Combined PII masking and OpenAI guardrails with custom open-source classifiers for prompt injection, adversarial inputs, and red-team attack patterns.

System outcome

Moved reliability work from manual spot checks toward trace-driven diagnosis, measurable evaluations, and layered security controls.

  • Langfuse
  • Custom evals
  • PII masking
  • Red teaming
03

Context and retrieval intelligence

Context Quality Engine

A context-selection layer that treats retrieval as one input to the agent’s working state—not the whole intelligence system.

  • Combined multi-query expansion, hybrid retrieval, reciprocal rank fusion, and reranking.
  • Evaluated evidence coverage and failure modes across sparse, high-specificity domain corpora.
  • Shaped compact, task-specific context under strict token budgets and domain constraints.

System outcome

Materially improved evidence coverage over keyword-only retrieval while preserving inspectability and source grounding.

  • Hybrid search
  • RRF
  • Reranking
  • Context selection
04

Data systems

Scalable Analytics Data Layer

A shared enriched data model that replaces repeated high-cost joins and aggregations in interactive analytics workloads.

  • Designed full and differential refresh strategies for evolving source data.
  • Moved expensive cross-collection enrichment out of the request path.
  • Introduced explicit freshness semantics for downstream dashboards.

System outcome

Reduced repeated computation and created a more scalable path for analytics growth.

  • MongoDB
  • ETL
  • Materialized data
  • CubeJS

02 / ENGINEERING APPROACH

Reliable agents are engineered, not prompted.

01

Runtime before prompt

Treat the model as one component. Engineer orchestration, tools, state transitions, validators, and response contracts around it.

02

Context is a budget

Assemble the smallest high-signal state for each decision—evidence, instructions, history, and tool affordances—not one oversized prompt.

03

Evals start with traces

Use real trajectories to find failure modes, convert them into repeatable scenarios, and measure every harness change against regressions.

04

Security is layered

Protect boundaries with PII controls, policy guardrails, adversarial classifiers, and red-team scenarios instead of relying on one filter.

03 / EXPERIENCE

Applied AI engineering with software-system depth.

JUL 2022 — PRESENT

Hyderabad, India

Senior Software Engineer

Regology India · Applied AI & platform engineering

  • Build production agentic systems for regulated workflows across orchestration, context, skills, validators, guardrails, and response layers.
  • Own reliability foundations spanning Langfuse tracing, a custom evaluation suite, PII controls, adversarial detection, and red-team scenarios.
  • Design context and search systems across Elasticsearch and MongoDB using hybrid ranking, fusion, reranking, and evidence-quality evaluation.
  • Improve data-platform scalability through materialized models and differential refresh pipelines, while contributing to technical design and engineering interviews.

04 / CAPABILITIES

Depth in agent systems. Breadth across the production stack.

01

Agent runtime & harness

The execution system around the model.

  • Agent orchestration
  • Context engineering
  • Harness engineering
  • Tool routing
  • Agent skills
  • Stateful workflows
  • Validators
02

Reliability & security

Evidence, controls, and feedback loops for production behavior.

  • Langfuse tracing
  • Custom evaluation suites
  • Regression testing
  • Guardrails
  • PII masking
  • Adversarial detection
  • Red-team testing
03

Context & search

High-signal evidence selection under finite attention.

  • Elasticsearch
  • Hybrid retrieval
  • Vector search
  • Multi-query expansion
  • Reciprocal rank fusion
  • Reranking
  • Source grounding
04

Backend, data & platform

The production foundations that keep AI systems useful.

  • Python
  • MongoDB
  • SQL
  • ETL pipelines
  • Materialized data models
  • AWS
  • Kubernetes
  • CI/CD

05 / RÉSUMÉ

Hari Krishna

Senior Software Engineer · Agentic AI systems

Profile

Senior Software Engineer building production agent runtimes across context, orchestration, evaluation, security, search, and data platforms.

Current experience

Regology India · July 2022 — Present

Education

BITS Pilani, Hyderabad Campus

Core focus

Agent runtime · Context engineering · Harness engineering · Reliability & security

This is a verified-content résumé snapshot. Contact details and the final résumé file will be added before public launch.

06 / CONTACT

Let’s build AI systems that hold up outside the demo.

Interested in senior software engineering opportunities focused on agent infrastructure, applied AI, reliability, and platform-scale problems.

Email Add addressLinkedIn Add profileGitHub Add profile