Hammad Farooq

AI Engineer specializing in Agentic LLM Systems, RAG, and Python/FastAPI backends.

Based in: Lahore, PakistanCurrently: Independent Data Scientist & Kaggle Competitor, Self-EmployedExperience: 4 years

Hire me

Hammad is an AI engineer with over 3 years of experience designing multi-step agent workflows and RAG systems for finance, healthcare, and legal-tech clients. He focuses on building verifiable, production-ready AI applications that prioritize grounded accuracy, structured outputs, and traceable audit logs.

Experience

Independent Data Scientist & Kaggle Competitor

Self-EmployedRemotefreelanceFeb 2025 – Present

Summary

Compete in research challenges such as ARC-AGI (an abstract reasoning benchmark), NVIDIA reasoning benchmarks, and Legal Information Retrieval, against thousands of teams.

What I did

  • Fine-tuned ByT5 transformer models for Akkadian translation, a low-resource language, handling data cleaning, tokenization, training, tuning, and evaluation myself.
  • Run side-by-side experiments under fixed GPU/TPU limits, track every one so results can be reproduced, and compare models on held-out data before trusting any improvement.
  • Conducted systematic experimentation under fixed compute budgets and reproducible evaluation for ARC-AGI and NVIDIA reasoning challenges.

Results

  • Placed 5th out of 79 teams in the Rainfall Probability Prediction Challenge.
  • Achieved a private ROC AUC of 0.9033 in the Rainfall Probability Prediction Challenge.
ByT5TransformersGPUTPUPython

Data Scientist & AI Developer

seKer AIRemoteworkMay 2023 – Present

Summary

Own AI subsystems for international clients in finance, real estate, legal-tech, and healthcare from the first architecture conversation through to a system running in production.

What I did

  • Build RAG-based document search and Q&A systems on FAISS and Chroma, with ingestion pipelines that pull documents from several sources so client teams can query their own data.
  • Build agents with tool calling and memory that handle multi-step questions, and add retrieval checks so the agent doesn't answer from thin or unrelated context.
  • Ship AI services as FastAPI APIs with the engineering team, then monitor latency, answer quality, and drift once they are live.
  • Add evaluation checks and guardrails to deployed LLM services so unsupported or off-topic answers are caught before they reach users.
  • Stay in direct contact with clients through the project, explaining progress, results, and trade-offs in plain language so they can make decisions quickly.
  • Owned AI subsystems from architecture and retrieval pipelines to FastAPI deployment and evaluation checks for answer quality and latency.

Results

  • Improved system reliability by adding source-grounded responses, checks to filter unsupported claims, and refusal handling for unsupported answers.
FAISSChromaFastAPIRAGLLMPythonJSON modeFunction callingn8nMakeZapier

AI Engineer (Part-Time)

IT Loopspart timeFeb 2025 – May 2026

Summary

Built chatbots and LLM-based automations for enterprise clients, working part-time alongside my main role.

What I did

  • Developed and maintained ML APIs in FastAPI and set up CI/CD on GitHub so every release was tested and versioned.
  • Turned client requirements into data pipelines and system designs together with the engineering team, which cut manual steps out of their processes.
  • Integrated models with CRMs and messaging tools using n8n, Make.com, and Zapier, so staff no longer had to move data around by hand.
  • Automated repetitive steps such as moving data between systems and triggering CRM or messaging actions.
  • Built CRM automations that extended into external actions following defined workflows.

Results

  • Cut manual steps out of client processes through automated data pipelines and system designs.
FastAPICI/CDGitHubn8nMake.comZapierCRM

Skills

Technical

RAG
FastAPIFastAPI
GPT-4o
PythonPython
Multi-step agents
LangChain
FAISS
Anthropic Claude API
Cursor
Claude Code
Pydantic
Ollama
PostgreSQLPostgreSQL
Chroma
Next.jsNext.js
zod
DockerDocker
TypeScriptTypeScript
ReactReact
Zapier
Multi-source retrieval orchestration
Model-agnostic orchestration
Retrieval validation
Make
CRM and messaging integration
SSE streaming
n8n
Agentic tool selection
Tool failure handling
Structured Outputs

Projects

Loophole Larry: Multi-Agent Legal AI Platform

AI Engineerprofessional
Built a multi-agent legal Q&A system for a data-sensitive profession that only answers when it can cite a source. Hybrid search finds the passages, entailment checks drop claims the sources don't support, and it says "I don't know" when the answer isn't in the corpus. Added refusal handling and full audit logs so every answer can be traced back to its sources, plus eligibility calculators for legal aid. Built a case-matching engine that ranks similar precedents with confidence bands (target of at least 85% precision), and a CPD course recommender with sensible defaults for users with thin profiles.
PythonLangChainGPT-4oRAGFastAPIPostgreSQLFAISSlegal-aid eligibility calculatorscase matching

Parley: AI Meeting Notes & Q&A

Full Stack AI Developerpersonal
Built a meeting intelligence product that turns transcripts into notes, action items, and highlights, with an "Ask" feature that answers questions across meetings and cites the exact transcript lines. Designed Ask as retrieval, then synthesis: Postgres full-text search ranks candidate meetings, a deterministic engine scores decisions, action items, and transcript lines, and Claude can write the final cited answer when an API key is set. Validated every endpoint with zod, returning clear 400/404/422 errors, built a transcript import pipeline (WebVTT, SRT, plain text), covered it with unit and integration tests, and deployed it live on Vercel with Neon Postgres.
Next.js 16TypeScriptPostgreSQLDrizzle ORMzodClaude API
Visit the site

RAG + Agents Chatbots with OpenAI and Local Ollama Models

AI Engineerpersonal
Built Q&A chatbots that switch between OpenAI models and open-source models running locally through Ollama, and served them through FastAPI with LangServe. Built a multi-source RAG chatbot with agents that choose between retrieval tools to answer multi-step questions, backed by embedding pipelines and advanced retrievers. Produced JSON-structured responses (question, retrieved context, answer) in a related RAG assistant, with SentenceTransformer embeddings and CLIP image embeddings for multimodal input.
PythonLangChainLangServeFastAPIOllamaFAISSChromaSentenceTransformersCLIPOpenAI

The Archive: Grounded Multi-Turn Document Q&A

Lead Developerpersonal
Built a document Q&A system where every answer shows the passages it came from, under a strict grounding prompt: the assistant says "nothing in the collection covers this" instead of filling gaps from memory. Handled multi-turn conversation state properly: history is read before the new question is stored, the last 6 turns go into the prompt, and each passage is labelled with its source, chunk, and relevance score. Built for correctness and failure cases: Pydantic contracts on every endpoint, exact scoped search over only the documents a user selects, SSE streaming with partial answers saved if the connection drops, and a clear error if the embedding model changes and the index no longer matches.
PythonFastAPIPydanticFAISSGPT-4oNext.js

RAG Assistant with Structured Responses

personal
Built a RAG assistant that returns a structured object containing the question, retrieved context, and answer instead of free-form text.
PydanticJSON-structured outputs

Education

BS in Data Science

Virtual University of PakistanData Science

Intermediate in Computer Science (ICS)

Punjab Group of CollegesComputer Science

Recognition

Awards

Certificate of Gratitude

Ivan Franko National University, Ukraine2026

Invited Speaker at AI Technologies Summer School 2026

5th Place - Rainfall Probability Prediction Challenge

Kaggle

Private ROC AUC 0.9033, 5th of 79 teams

Certifications

Microsoft AI & ML Engineering Professional Certificate

Microsoft

IBM Data Science Professional Certificate

IBM

Meta Full Stack Developer: Front-End & Back-End from Scratch

Meta

Google Advanced Data Analytics Professional Certificate

Google

Statistics with Python Specialization

University of Michigan

Social impact & activities

Participant

LabLab.ai Hackathonsother

Built working AI solutions under tight time and resource limits.

Building Real-World AI Apps with LLMs and RAG

AI Technologies Summer School 2026conference talk2026

Gave a lecture at an event with 70 speakers from 21 countries and 1,335 participants from 60 countries.

Hire HammadReach out about a role, a contract or a conversation.For recruiters

Hammad's twin is AI, it can make mistakes.

More people on Kanvis

Browse more peopleCreate your own Kanvis