Skip to content
SK
Currently accepting projects

Open to selected freelance AI/ML projects

Currently working full-time as an AI/ML Engineer, and taking on a small number of freelance engagements alongside it - which means I am selective, and honest about timelines before we start.

Capacity

Part-time, limited engagements

Response time

Usually within 24 hours

I work with

Startups, Product teams, Research groups, Agencies

Services

What I can build for you

Every service listed here maps to something I have actually shipped — the note under each one says what backs it up.

Custom RAG Systems

Retrieval over your own documents, data or knowledge base - including the cases where naive chunk retrieval falls apart and the domain needs to be modeled as a graph.

  • Ingestion & chunking pipeline
  • Hybrid semantic + BM25 retrieval
  • Graph RAG where relationships matter
  • Retrieval evaluation harness
Backed by: Built and evaluated hybrid-search and Graph RAG pipelines over clinical trial, drug and disease data on client work.

AI Agents & Agentic Workflows

Multi-agent systems that do real work against real tools - orchestrator/sub-agent architectures rather than one prompt asked to do everything.

  • LangGraph / CrewAI orchestration
  • Specialized sub-agents per concern
  • Agent memory design
  • Failure handling and retries
Backed by: Developed 10+ specialized LangGraph agents in an orchestrator/sub-agent architecture on enterprise client work.

MCP Tool Servers & Integrations

Standardized tool access for your agents, so adding the next integration does not mean inventing another bespoke protocol.

  • MCP servers (SSE and stdio)
  • Tool schema design
  • Database and API tool nodes
  • Schema introspection for dynamic SQL
Backed by: Designed MCP servers with SSE and stdio transports, including schema-introspection and query-execution tools.

LLM Chatbots & Assistants

Assistants grounded in your content that answer from what you actually published, and say so when they do not know.

  • Grounded question answering
  • Conversation memory
  • Fallback and refusal behavior
  • Web or API delivery
Backed by: Built research-facing Q&A chatbots over retrieval layers, and conversational interfaces over automation pipelines.

Natural Language to SQL / Query

Let non-engineers ask your database questions in plain English, against the live schema rather than a stale snapshot.

  • Schema introspection layer
  • Query generation and validation
  • Governance rules as agent context
  • Result explanation
Backed by: Built natural-language-to-SQL workflows over introspected schemas, and NL-to-Cypher/Gremlin translation for graph databases.

Guardrails & AI Safety Review

Input and output screening for systems that touch regulated or personal data - and a look at what your agent does when someone tries to talk it out of its instructions.

  • Input / output guardrail layer
  • PII / PHI screening
  • Prompt-injection resistance
  • Policy-aligned refusal behavior
Backed by: Implemented guardrails covering PII/PHI, jailbreaks, harm, social bias, profanity and prompt injection alongside IBM Guardrails.

LLM Evaluation Harnesses

A way to tell whether your last prompt change helped or hurt, instead of arguing about sample outputs.

  • Labeled evaluation dataset
  • LLM-as-judge scoring
  • Faithfulness & relevancy metrics
  • Regression tracking between iterations
Backed by: Built an LLM-as-judge evaluation harness scoring responses against a labeled dataset, and a shadow-evaluation harness with Bayesian threshold tuning.

Cost & Latency Optimization

Routing, caching and model-tier decisions for systems where the inference bill has started to matter.

  • Query complexity routing
  • Cost-tiered model dispatch
  • Response caching
  • Cost and routing dashboard
Backed by: Built an adaptive routing gateway with a fine-tuned complexity classifier, cost-tiered dispatch, Redis caching and Optuna threshold optimization.

How it works

From first message to shipped work

01

Tell me what you are building

Send the problem, not the spec. What the system needs to do, who uses it, and what breaks today.

02

Scope call

A short call to pin down constraints, data access and what success actually looks like.

03

Proposal & timeline

A written scope with milestones and an honest timeline that accounts for my full-time commitments.

04

Build & iterate

Working increments you can see, with evaluation in place so quality is measured rather than asserted.

Project inquiry

Tell me what you are building

Send the problem, not the spec. I read every inquiry personally.

What happens next

I reply within about a day — usually with a couple of questions before anything else. If it is not a fit, I will say so quickly rather than leave you waiting.

StartupsProduct teamsResearch groupsAgencies

Optional

The problem, who uses it, and what breaks today. Specifics help more than a spec.

Optional

Optional

Optional — constraints, data access, existing stack, deadlines.

Or email me at Ksaikiran129@gmail.com