Skip to main content

Umarfarook Gurramkonda · applied AI/ML · HypeOn AI, Bengaluru

I measure what I ship.

Multi-agent LLM systems, retrieval, and natural-language interfaces over data. Every system ships with an eval harness, a cost cap, and numbers you can rerun yourself.

$ mail umarfarook0yt@gmail.com
GitHub ↗LinkedIn ↗

01 / evidence

The metrics are the imagery.

0%

strict 7-field extraction from raw quote emails

fig 1.1 · Cargo Concierge

0.000

YOLOv8n mAP@0.5, license-plate detection

fig 1.2 · street-view-plate-blurring

~$0.000

cost per agent quote, end to end

fig 1.3 · Cargo Concierge

+0 pts

from the instruction block alone

Cargo ablation

0

trust dimensions per agent answer

TrustBench

0 MB

mandatory query cost cap

mcp-bigquery-evals

~0.0 ms

detector inference, 3.0M params

YOLOv8n

02 / work

Case studies with harnesses.

  1. 01Cargo Conciergeagentic product80% strict extraction · ~$0.002/quote2025 ↗
  2. 02mcp-bigquery-evalsopen-source infra · PyPI100 MB cost cap · 7 MCP tools2025 ↗
  3. 03TrustBenchevaluation system6 trust dimensions · calibrated judges2025 ↗
  4. 04RAG Document QAretrieval systemRecall@K + MRR harness2025 ↗
  5. 05License Plate Privacy Blurringcomputer visionmAP@0.5 0.782 · ~4.1 ms2026 ↗
  6. 06IPL Franchise Analyticsdata analysis1,095 matches · 17 seasons2026 ↗
  7. 07Shorts Performance PredictionML analysis · negative resultp = 0.955 · no signal, published anyway2026 ↗
every repo → github.com/Umarfarook1 ↗

03 / method

Route, ground, constrain, measure.

[01]▸ orchestrateRoute intent, retrieve context, call tools, stream progress, fail over cleanly across providers.proof · Cargo-Concierge
[02]▸ groundHybrid search, reranking, metadata filters, schema discovery, natural-language access to warehouses.proof · rag-document-qa
[03]▸ constrainStructured outputs, validation, policy gates, deterministic business logic, cost-aware execution.proof · mcp-bigquery-evals
[04]▸ measureGolden sets, regression slices, calibrated judges, latency and cost traces, failure attribution.proof · trustbench

04 / checkpoints

Checkpoints on the run.

  1. 2024 → now

    ML Engineer

    HypeOn AI

    Production AI for D2C trend intelligence: multi-stage orchestration, RAG, NL-to-SQL over BigQuery with cost guardrails, observable deployments on GCP.

    LangChain · FastAPI · BigQuery · Cloud Run · Gemini

  2. 2024 → 2025

    Freelance ML / AI Engineer

    Independent

    AI-assisted inventory system for a retail client: invoice extraction, demand forecasting, real-time stock alerts, operational dashboard.

    Python · scikit-learn · OpenAI · SQL

  3. 2024

    Software Engineering Intern

    Synclovis Systems

    FastAPI and Flask services for an LLM healthcare assistant over 500+ clinical documents, LangChain + FAISS retrieval, AWS deployment, guardrails that cut hallucinations on out-of-scope queries.

    FastAPI · LangChain · FAISS · AWS

education

B.Tech · Computer Science · 2024

K.S.R.M College of Engineering · 8.14 / 10

certifications

  • Oracle OCI Data Science Professional · 2025
  • Oracle OCI AI Foundations Associate · 2025
  • Azure AI Fundamentals · AWS Cloud Foundations

05 / working set

models + agents

backend + data

delivery

interface

06 / about

Umarfarook Gurramkonda
fig. 06 · the source image

Most of my time goes to the layer around the model.

Deciding when a model earns its place, shaping what goes in and out, and checking the result still holds under real traffic. Evaluation, cost discipline, failure handling: the unglamorous work that makes a system dependable. I want research and ML engineering roles where that judgment counts as much as model choice.

Umarfarook.

07 / contact

Write to me.

$ mail umarfarook0yt@gmail.com

I read my own inbox and reply fast. Research and ML engineering roles where evaluation and reliability count.