Monish Gosar

Monish Gosar, a forward-deployed engineer building agentic systems end to end.

I design and ship production agentic systems, from discovery to multi-agent architecture, serving, and evaluation. The work is systems that understand the domain they run in: purpose-built skills, grounded tool calls, and multi-turn analysis that keeps its context.

A data-science background sits under that. Every workflow is only as good as the data it is allowed to trust.

I’m open to conversations. Reach me by email, on GitHub, LinkedIn, or X.

01 / Experience

Where I’ve worked

  1. Indilabs.aiCurrent

    Forward Deployed Engineer (Founding Team) · Remote · Nov 2024 – Present

    Embedded with operating teams to turn manual workflows into agentic systems, then stayed through adoption.

    • Designed a co-analyst framework that turns recurring analytical workflows into reusable skill cards: fetching live data, running domain analyses, and producing structured, decision-ready outputs.
    • Architected a multi-agent analytics platform on Azure OpenAI + OpenAI Agents SDK with schema-constrained tool calling and live data services: auditable answers in under 15 seconds, cutting reporting effort 40%.
    • Owned the LLM serving layer: token-budget management, rate limiting, latency reduction; plus a production ML inference layer (delinquency, propensity, forecasting) behind low-latency REST APIs across 21+ products.
    • Built a custom agent runtime with parallel tool execution, structured outputs, Langfuse tracing, and LLM-as-a-judge evaluation for factuality, completeness, and bias.
    • Shipped a contactless settlement product with NPV/DCF-optimized offers, propensity scoring, and an AI negotiation layer: +12% settlement acceptance.

    How it fits togetherAgents, the serving layer, and the models sit on one path: live data in, a traced answer out, with the token budget enforced before the call.

    Azure OpenAI · OpenAI Agents SDK · Langfuse · FastAPI · Machine Learning

  2. RE Journal

    Data Scientist · Remote · May 2024 – Aug 2024

    Built the analytics behind buyer profiling, from raw filings to a dashboard a team could use.

    • Segmented 1,000+ users into 3 behavioral clusters via K-Means across 6 features for buyer profiling insights.
    • Built analytics pipeline: scraped 150K+ RERA records (Selenium, BeautifulSoup), modeled SQL database, delivered Power BI & Streamlit dashboards.

    How it fits togetherRERA records land in SQL, then in the Power BI and Streamlit dashboards.

    Python · Selenium · SQL · Power BI · Streamlit

02 / Projects

Things I’ve built

02

Folio — Zero-Trust RAG

Policy-driven RAG where access is decided in code, never by the model. Each vector chunk carries its tenant and role ACL, so unauthorized documents never reach the LLM.

PythonFastAPIQdrantAzure OpenAIReact
Read write-up ↗github ↗

03 / Writing

The engineering, written up

15 min

AI's Real Fight Is Over Compute, Money, and Power

AI is becoming a heavy industry, and the race for chips, electricity, and capital could reshape markets, governments, and who holds power.

1 min

How to Build a PR Review Agent

How I built a PR review agent that scans every changed file, ranks risk, pulls the right context, and keeps track of what was actually reviewed.

5 min

A note on building agents without rewriting your product for CRUD based apps.

Every mature product has a layer of plain database functions. That layer is where your product actually lives.

18 min

Humanity's Cost to AGI

We're living through what might be the most significant financial event of our generation as AGI development valuations skyrocket.

04 / Contact

Building something? Let’s talk.

Email me at monish.emailbox@gmail.com, or find the work on GitHub and LinkedIn.

Monish Gosar