TradePulse
Backtest strategies with next-bar execution and trading costs included.
Keep the future out of the backtest.
Decide on bar i, fill at bar i+1's open; a canary test fails the build if any look-ahead leaks into a result.
Mubin Attar Applied AI Engineer
AI systems,built for the real world.
I build the software around the model: schema-aware agents, dependable APIs, and interfaces people can use. Here are the decisions, trade-offs, and tests behind the work.
Ahmedabad, IndiaSoftware since 2022AI focus since 2024
Selected system / 01
DBWhisper
The model proposes. The validator decides.
Proposed SQLPostgreSQL
SELECT created_at, amount
FROM orders
ORDER BY created_at DESC
LIMIT 10;orders is present in the enrolled schema.
Validator checks
- Single statementPass
- Read-only statementPass
- Enrolled schemaMatched
Eligible for read-only execution
This example satisfies the illustrated checks. Database permissions, a timeout, and a row cap remain separate execution controls.
18/22 exact golden queries · 4/4 unsafe test prompts refused
Separate evaluation · 8 Jul 2026 · Method & limits
Selected work
More systems, different problems.
Predictive ML, financial research, and multi-user AI. Explore the implementation and the decisions behind each product.
CrownWager
Turn a validated model into odds, edge, and a graded record.
Make the model evaluation repeatable.
65.2% ± 0.8%model accuracy5-fold stratified cross-validation on 15,115 NBA games; base home-win rate is 57.5%, so a real ~8-point edge (ROC-AUC 0.685).
LLM Studio
A multi-user LLM platform with tenancy enforced below the route.
Enforce ownership in the data layer.
Queries for chats and history are scoped to the authenticated owner in the repository layer, alongside route-level authentication.
The engineer
The engineer behind the work.
Mubin Attar
Applied AI Engineer
Ahmedabad, India
I'm an applied ai engineer with 3+ years in software engineering and a production-AI focus since 2024. I work across the system: data pipelines, APIs, retrieval, model integration, and the interface people use.
Professional experience
2024 – present
AI/ML Engineer
Sevina Technologies
Production healthcare-AI automation — clinical-compliance and reimbursement pipelines (MDS/PDPM), LLM document analysis with multi-provider routing and RAG, and eligibility-verification services, Dockerized and CI'd.
2022 – 2024
Junior Python Developer
Linescripts Software
End-to-end web applications and RESTful services — a Hospital Management System, e-commerce platforms, and third-party API integrations, across Python, Django/DRF, and a JavaScript frontend.
From the notebook
Thinking in public.
Selected writing
All writing- A validator is not a better promptWhy DBWhisper gates every generated query with deterministic code instead of a stricter system prompt — 'less often' is not a safety property.essay ·
- Trust is not a safety modelWhy I gate an LLM's SQL behind a deterministic, fail-closed validator instead of trusting the model to behave.essay ·
- Look-ahead-free by constructionThe most common way a backtest lies to you is look-ahead bias. Here's how I designed it out of TradePulse instead of hoping to catch it later.essay ·
Currently exploring
What I'm doing nowGolden-query evals for text-to-SQL
Execution accuracy against a real database, not string-matching against a reference query.
Model Context Protocol for typed tool-use
Tool contracts as a protocol between agents, rather than bespoke glue per integration.
Long context vs. retrieval
When a bigger window still loses to a small, scoped prompt — and how to tell in advance.
Building an AI product that needs to work in production?
Let's discuss the system, the failure modes, and how success will be measured.
sk.mubinattar@gmail.comOpen to senior AI engineering and applied AI roles — remote or Ahmedabad, India