You will get a production RAG chatbot over your documents with citations and search
Rising Talent
Rising Talent
Project details
I build RAG chatbots that don't hallucinate, don't cite the wrong source, and don't break at 1,000 documents.
Most RAG demos work on 10 PDFs. Production RAG works on 5,000 — that's an engineering problem, not a prompt problem.
What sets this apart:
→ Hybrid retrieval. Semantic search alone misses keyword matches. I combine vector search + keyword (BM25) + re-ranking for measurably better citations.
→ Production scale. Shipped on HomeScanr (Claude + pgvector, live on App Store + Play Store) and TPE-ATS BACON (NestJS multi-agent over enterprise docs).
→ Citations that actually point to source pages, not chunk IDs. Buyers can verify every answer.
→ Eval harness included. Measure retrieval quality on your test set, not just hope it works.
→ Chunking strategy tuned to YOUR data — long-form, Q&A, tables, code — not generic 512-token windows.
→ You own the code. Source delivered to your GitHub, deployment guide included on Standard and Advanced tiers.
8,700+ hours on Upwork. 100% Job Success. Available 15-25 hrs/week. Karachi (UTC+5).
Tech: Claude API, OpenAI, LangChain, LlamaIndex, Pinecone, pgvector, Weaviate, Vercel AI SDK, Next.js, Node.js, Python.
Most RAG demos work on 10 PDFs. Production RAG works on 5,000 — that's an engineering problem, not a prompt problem.
What sets this apart:
→ Hybrid retrieval. Semantic search alone misses keyword matches. I combine vector search + keyword (BM25) + re-ranking for measurably better citations.
→ Production scale. Shipped on HomeScanr (Claude + pgvector, live on App Store + Play Store) and TPE-ATS BACON (NestJS multi-agent over enterprise docs).
→ Citations that actually point to source pages, not chunk IDs. Buyers can verify every answer.
→ Eval harness included. Measure retrieval quality on your test set, not just hope it works.
→ Chunking strategy tuned to YOUR data — long-form, Q&A, tables, code — not generic 512-token windows.
→ You own the code. Source delivered to your GitHub, deployment guide included on Standard and Advanced tiers.
8,700+ hours on Upwork. 100% Job Success. Available 15-25 hrs/week. Karachi (UTC+5).
Tech: Claude API, OpenAI, LangChain, LlamaIndex, Pinecone, pgvector, Weaviate, Vercel AI SDK, Next.js, Node.js, Python.
Programming Languages
JavaScript, Python, TypeScriptCoding Expertise
Cross Browser & Device Compatibility, Performance Optimization, SecurityWhat's included
| Service Tiers |
Starter
$400
|
Standard
$1,200
|
Advanced
$3,500
|
|---|---|---|---|
| Delivery Time | 7 days | 14 days | 21 days |
Number of Revisions | 1 | 2 | 3 |
Design Customization | - | - | - |
Content Upload | - | - | - |
Responsive Design | - | - | - |
Source Code | - | - | - |
Frequently asked questions
2 reviews
(2)
(0)
(0)
(0)
(0)
This project doesn't have any reviews.
HM
Hansel M.
Aug 16, 2022
Website developer for video forwarder
Hasham went out of his way to make sure I was completely satisfied with the website (both on mobile and desktop), he was also able to figure out some bugs that my code had very quickly. He is friendly, speaks English perfectly and his prices are not expensive. I would definitely recommend Hasham for any react work or website development.
HM
Hansel M.
Aug 16, 2022
Website development with web3
Hasham went out of his way to make sure I was completely satisfied with the website (both on mobile and desktop) m, he was also able to figure out some bugs that my code had very quickly. He is friendly, speaks English perfectly and his prices are not expensive. I would definitely recommend Hasham for any react work or website development.
About Muhammad Hasham
AI Agent, RAG, and AI SaaS Engineer - Claude, Next.js, Production
100%
Job Success
Karachi, Pakistan - 5:53 am local time
8,700+ hours · 100% Job Success · Rising Talent · Available 15-25 hrs/week
I use Claude Code daily on live production work — most "AI developers" can't say that.
WHAT I SHIP:
→ AI Agents
Multi-tool, memory, Claude or OpenAI. Production-deployed with observability + error handling. Real agents that handle real users, not demo scripts.
→ RAG Chatbots
Hybrid retrieval (semantic + keyword), re-ranking, citations, eval harness. Production scale, not 10-PDF demos.
→ AI SaaS MVPs (Code-First)
Hand-built Next.js + NestJS + Stripe + LLM backend. Real codebase you own — not a Lovable/Bolt shell you'll outgrow.
→ Inbound Voice + WhatsApp AI Bots
Vapi, Retell, WhatsApp Cloud API with Meta signature verification. Customer support and appointment booking flows.
→ Senior Production Debug
Next.js, Node, NestJS — root-cause analysis, Loom walkthroughs, regression tests.
LIVE BUILDS:
• HomeScanr (Claude + OpenAI, App Store + Play Store)
• (NestJS migration from Clarion legacy)
• Matchportalen (multi-tenant SaaS, 99.9% uptime)
• ProposalGenie (500+ subscribers, win rates tripled)
• DripMap (10K+ MAU, MVP shipped in 5 weeks)
TECH:
Next.js · NestJS · TypeScript · Claude API · OpenAI · LangChain · LlamaIndex · Pinecone · pgvector · PostgreSQL · AWS · Stripe · Docker
WHAT CLIENTS SAY:
"He does great work and puts in a lot of effort to make sure the results are as good as they can get."
"He fixed the bug blazingly fast, super nice with explanation of what was the bug."
If you need AI integrated into your product, or a SaaS built right the first time — message me.
— Hasham
Karachi (UTC+5), available on US/EU morning calls
Steps for completing your project
After purchasing the project, send requirements so Muhammad Hasham can start the project.
Delivery time starts when Muhammad Hasham receives requirements from you.
Muhammad Hasham works on your project following the steps below.
Revisions may occur after the delivery date.
Step One
Kickoff call to confirm document sources, use case, and retrieval requirements
Step Two
Document ingestion pipeline built with chunking strategy tuned to your content