SQL · Python · Power BI · Tableau · Excel | RAG · AI Agents & MCP · LLM Evaluation & Deployment
I turn data into decisions, and then I build the AI that keeps answering the next question.
- As an analyst: 5+ years in data analytics and business intelligence across FinTech, Banking and SaaS (Axis Bank, AI and analytics startups). I build revenue pipelines, A/B tests, forecasts and executive dashboards for ARR, MRR and churn, starting from the business metric and proving results with numbers.
- As an AI engineer: I build production-grade AI systems: retrieval-augmented generation with measurable accuracy, tool-using agents, and the evaluation and deployment pipelines that make them trustworthy.
Jump to: 📊 Analyst & BI work · 🤖 AI engineering work
Selected impact:
- 📉 80% fewer reporting discrepancies with automated data validation for revenue-critical leadership reporting
- ⚡ 66% faster compliance reporting queries at Axis Bank (15 min → under 5), with 99.9% data accuracy
- ⏱️ Manual reporting cut from hours to under 45 minutes through automated workflows
- 🏦 ETL pipelines processing several million records daily across core banking and loan systems at 99% reliability
Business questions answered with SQL, statistics and dashboards, grouped by the analyst role they fit.
| Role fit | Project | Business result |
|---|---|---|
| Business / Business Systems Analyst | Requirements-to-Insight Copilot · ▶ demo | Traced 6,534 late deliveries (6.8% of 96k orders) to 3 root causes; wrote the BRD, user stories and to-be process; late seller handoffs carry 3.9x late risk, 83 sellers cause half |
| Management / Operations Analyst | Service Operations Bottleneck Analyzer · ▶ demo | 281k NYC 311 requests: weekend staffing gap makes Fri-Sat filings 2.4x slower (~439 requests wait for Monday weekly); slowest 10% hold 77% of request-days; staffing scenarios |
| Project / Program Analyst (PMO) | PMO Portfolio Health · ▶ demo | 895 real milestones from 17 large software programmes: only 32.9% land on the day but 70.6% within a week; RAG board, P85 reference-class forecasts and an Excel RAID log; found 19 close-outs inflating overdue counts |
| Healthcare / Clinical Data Analyst | Hospital Readmissions Analyzer · ▶ demo | CMS readmission data for 3,055 hospitals: 83.2% are above expected somewhere, but 219 hospitals hold half of the 12,245 excess readmissions; separates hospital-wide from single-service problems (p < 0.001) and ranks a call list; six-page Power BI dashboard with DAX checks against the Python figures |
| Supply Chain / Logistics Analyst | Supply Chain Shipment Analyzer · ▶ demo · 📊 Tableau | USAID shipments of medicines to 43 countries: $259M of stock arrived late; warehouse route late 17.2% vs 5.3% direct (6.1x odds like for like, p = 0.001); air is 62.5% of freight; moving heavy early orders to sea saves $1.73M; two-dashboard Tableau Public viz generated as code |
| Economics / Economic Analyst | India Inflation Analyzer · ▶ demo | Official MoSPI CPI, 2014-2026: inflation doubled to 5.54% in 2026 and food supplies 48.9% of it; tested pass-through shows headline returns to core (coefficient 1.03, t = 4.2) while food does not spread, but core is now 4.6%, above target; seven-page Power BI dashboard with 14 DAX-vs-Python checks |
| Product / Marketing Analyst | Netflix content A/B test | 7.17% lift in engagement (p < 0.001) from longer previews; power analysis and t-tests |
| Marketing / Growth Analyst | Marketing mix model | Across $2.5M ad spend, podcasts had the highest marginal ROI; recommended a 15% budget reallocation |
| Operations / Supply Chain Analyst | Last-mile logistics bottleneck | Isolated a $192k/month recovery cost at one hub; showed the delay was operational, not geographic |
| Data / Customer Analyst | SaaS churn & retention engine | 82% recall churn model; support latency identified as the main churn driver |
| People / HR Analyst | Employee retention prediction | Attrition model with recall > 0.80; workload and burnout identified as the main drivers |
| Financial Analyst | AI Due-Diligence Analyst | Data-room review with deterministic financial tools; found 7/7 planted red flags, 0 false alarms |
| Risk Analyst | Real-time fraud detection | Flags risky transactions in under 50 ms with SHAP reason codes for every decision |
| BI Analyst | Power BI · Tableau · Excel / Sheets | Data modelling, DAX, Power Query ETL, Power Pivot star schemas, VBA automation |
| Project | What it does | Proof |
|---|---|---|
| India Inflation Analyzer · ▶ live demo | Inflation assistant over 9 governed tools (OpenAI function schema, offline planner): maps everyday words to official CPI components ("dal", "gold", "petrol"), parses years, writes a one-page monthly brief, and a grounding check rejects any number not returned by a tool. The headline statistical claim is re-tested in CI. | 600/600 tampered answers caught, 41/41 grounded; routing 26/26 tuned, 12/14 held-out reported as is; CI-gated |
| Supply Chain Shipment Analyzer · ▶ live demo | Supply-chain assistant over 8 governed tools (OpenAI function schema, offline planner): country and vendor resolution, scenario thresholds parsed from the question ("air over 2 t planned 150 days ahead"), one-page country briefs, and a grounding check that rejects any number not returned by a tool. | 736/736 tampered answers caught, 81/81 grounded; routing 24/24 tuned, 11/14 held-out reported as is; CI-gated |
| Hospital Readmissions Analyzer · ▶ live demo | Quality assistant for hospital readmissions: tool-calling planner (OpenAI function schema, offline fallback) over 6 governed tools, one-page hospital briefs, and a grounding check that rejects any number not returned by a tool. Statistical claims are tested before they are written. | 631/631 tampered answers caught, 98/98 grounded; tool routing 24/24 tuned, 13/14 held-out reported as is; CI-gated |
| PMO Portfolio Health · ▶ live demo | Tool-calling status agent (OpenAI function schema, offline planner fallback) that answers PMO questions and writes the weekly status report from governed metrics; a grounding guard rejects any number or date not returned by a tool. Out-of-time slip-risk model picks the simplest feature set that holds up. | 460/460 tampered answers caught; held-out tool routing 13/14 reported, not tuned; slip model AUC 0.85 out of time; CI-gated |
| Service Operations Bottleneck Analyzer · ▶ live demo | Weekly ops-brief agent: code computes every number, the LLM only phrases it, and a unit-aware grounding guard rejects any number, unit, date or agency not in the facts (one retry, then template fallback). | 212/212 tampered briefs caught, 0 false alarms; 14/14 LLM-style drafts judged correctly; CI-gated |
| Requirements-to-Insight Copilot · ▶ live demo | Copilot that answers stakeholder questions from a BRD, policies and a delivery data warehouse: governed metric layer first, guarded text-to-SQL as fallback, two-wall DuckDB sandbox, runs offline or on any OpenAI-compatible model. | 30/30 red-team SQL blocked; held-out eval reported next to the tuned set (95% routing); 73 tests in CI |
| Enterprise Knowledge Copilot · ▶ live demo | Answers employee questions from company documents with citations. Hybrid search (Qdrant + BM25), cross-encoder reranking, role-based access control and prompt-injection guardrails. | 44-case eval suite gated in CI: 100% top-1 retrieval, 0% access-control leaks, 100% correct abstention |
| AI Due-Diligence Analyst · ▶ live demo | Multi-agent system (LangGraph) that reviews a startup's data room and writes an investment memo: planner, researcher, analyst with deterministic financial tools, fact-checking critic, and human approval before the memo. | Finds 7/7 planted red flags with 0 false alarms; critic rejects fabricated claims; CI-gated evals |
| mcp-sql-analyst · MCP server | Secure, read-only SQL MCP server that lets Claude, Cursor or any MCP client analyse CSV / Parquet data: SQL guard + read-only locked DuckDB sandbox, PII masking, audit log, SaaS-metrics prompt. | 52/52 red-team attacks blocked in CI; 95 tests incl. stdio end-to-end |
| llm-gateway · ▶ live demo · 🐳 image | OpenAI-compatible LLM gateway (FastAPI) in front of Groq, Gemini, OpenRouter and Ollama: retries, fallback, circuit breakers, exact + semantic cache, virtual keys with budgets and rate limits, per-request cost tracking, Prometheus metrics. Chaos benchmarks with fault injection. | Under a 20% provider failure rate: 81.9% → 100% success; breaker cuts wasted calls 3,000 → 1,029; cache cuts FAQ spend 87% |
| agent-trading-firm · ▶ live dashboard | Multi-agent AI trading firm: six floors (scanner, research, strategy, testing, risk, control) paper-trade 20 crypto coins every hour on live Coinbase data. Code sets every number; the LLM (via my llm-gateway) writes the research and thesis. Same code runs the backtest and the live loop. | Honest 2-year backtest with ablations: risk floor halves max drawdown (−25% vs −45%); entry rule shows no edge after costs; anonymised test showed LLM vetoes removed winning trades, so its opinion is now advisory |
| bank-txn-parser-qlora · ▶ live demo · 🤗 model | QLoRA fine-tune of Qwen2.5-1.5B on a free Kaggle T4 that turns Indian bank SMS alerts and statement lines (UPI, NEFT, card, NACH …) into JSON. Regex and few-shot baselines, held-out merchants and formats, 8-bit GGUF for CPU. | On unseen merchants/formats: 72% exact match vs 32% regex, 17% few-shot; 98% on seen formats |
| YouTube AI Research Agent | Durable AI agent that indexes large YouTube playlists (TED-Ed) and turns long-form video into searchable insights. | Python agent pipeline |
| Real-Time Fraud Detection | Flags high-risk transactions in real time and explains every decision. | Random Forest · FastAPI · SHAP explainability |
| Role | Organisation | Period |
|---|---|---|
| Business Intelligence & Analytics Consultant | Humantic AI · EarthSense Labs · DataNeuron (AI, SaaS, Customer Intelligence) | Jun 2024 – Present |
| Deputy Manager | Axis Bank | Feb 2024 – May 2024 |
| Data Analytics Consultant | Credgenics · Intello Labs · ParallelDots (FinTech, Agritech AI, Retail AI) | Sep 2022 – May 2023 |
| Career Development Officer | Jaro Education | Sep 2021 – Jul 2022 |
| Data Analyst Intern | Cloudshope Technologies | May 2020 – Dec 2020 |
| Freelance Data Analyst | LUMIQ · ClickPost · Wiom (FinTech, Logistics, Telecom) | Jan 2018 – Jul 2020 |
Full details on LinkedIn.
- PGDM, Banking & Relationship Management, NIIT University, Neemrana (2023–2024) · CGPA 8.50
- MBA, Finance & HR, Maharaja Agrasen Himalayan Garhwal University (2021–2023) · Grade A
- B.Tech, Computer Science & Engineering, Uttar Pradesh Technical University (2017–2021) · First Division with Honours
Data & BI: SQL (SQL Server, MySQL, MongoDB), pandas, NumPy, scikit-learn, R, statistics and A/B testing, SHAP, time-series forecasting, Power BI (DAX, Power Query), Tableau, Excel (Power Pivot, VBA)
AI engineering: LLMs (OpenAI-compatible APIs, Ollama), RAG, embeddings, rerankers, Qdrant, LangGraph / agent frameworks, MCP, QLoRA fine-tuning, LLM evaluation, n8n automation, voice agents
Engineering: Python, FastAPI, ETL pipelines, Streamlit, Docker, GitHub Actions CI/CD, pytest
Domain: SaaS revenue metrics (ARR, MRR, churn, retention), banking and FinTech reporting, Agile/Scrum
AI engineering tracks (Udemy, 2026):
- AI Engineer Core Track: LLM Engineering, RAG, QLoRA, Agents (33.5 h)
- AI Engineer Agentic Track: The Complete Agent & MCP Course (21 h)
- AI Engineer Production Track: Deploy LLMs & Agents at Scale (18.5 h)
- AI Coder: Complete Claude Code & Coding Agents (16.5 h)
- AI Builder: Agents, Voice Agents & Automations in n8n (14.5 h)
- AI Leader: Generative AI & Agentic AI for Leaders & Founders (4 h)
Data & AI certifications:
- Databricks: Generative AI Fundamentals (2026)
- MongoDB: Vector Search (2026)
- Google: Data Analytics, Advanced Data Analytics and Business Intelligence Professional Certificates (2025)
- NPTEL+ / IIT Madras: Decoding the EU AI Act (2025)
- Analyst series: one industry project per analyst field (marketing and finance next; business systems, operations, PMO, healthcare, supply chain and economics done), each with a dashboard, a written recommendation and an AI assistant on top.
- AI engineering series: production-style projects, each with a live demo, an evaluation suite and CI: bilingual voice receptionist · AI code-review agent · document-intake automation.
Hiring an analyst or an AI engineer? I'd love to talk: LinkedIn · tanzilali.rizvi@gmail.com