Product Manager @ HashedIn by Deloitte | AI B2B| RAG, LLMs, Claude API in production | $1.2M ARR roadmap

Bengaluru · Hybrid

Bengaluru · Hybrid







Bachelor of Engineering - BE, Computer Science
Rungta College of Engineering & Technology Kohka-Kurud Bhilai
Join Reema and 10k+ other professionals on
Tal is the app where bosses meet engineers & builders directly, without middlemen
Download Tal BossProduct Manager @ HashedIn by Deloitte | AI B2B| RAG, LLMs, Claude API in production | $1.2M ARR roadmap
Twelve percent of our LLM feature's answers were wrong. We shipped at three percent — and most of the fix was product, not model. I'm a Product Manager at HashedIn by Deloitte, where I own the roadmap for an enterprise B2B SaaS module contributing $1.2M ARR. Six releases across four quarters, a seven-person cross-functional pod, a backlog groomed across 18 sprints, and on-time delivery moved from 72% to 90%. The AI work is the part I care most about: a Retrieval-Augmented Generation insights feature on the Claude API, with an evaluation framework across three prompt variants defined with client stakeholders. Hallucination-related defects dropped from ~12% to ~3% before go-live. I also run funnel and cohort analysis in Amplitude with custom SQL — one round of drop-off fixes drove a ~12% activation lift across two enterprise accounts. I started as a frontend engineer: two years in Angular and React on apps serving 10,000+ daily active users. I moved into product because I kept rewriting the requirements before building them. Those years are why I'd rather read the query than ask for the dashboard, and why I write acceptance criteria engineers don't have to translate. Right now I'm building Compass — an in-house delivery governance platform I took from zero. 18 discovery interviews, 3 shadowed audit cycles, a PRD and phased roadmap, leadership buy-in for a two-engineer pod. It now serves 200+ delivery managers across 30+ client engagements at ~70% weekly active usage. The conversations I want: AI evaluation in production, enterprise onboarding and activation, and 0-to-1 discovery inside organizations that don't think of themselves as product companies. If you're shipping an LLM feature and haven't yet defined how you'll measure it failing, send me your eval plan — I'll give you notes on it this week.
Bengaluru
claude
cursor
gemini
github copilot
lovable
notebooklm
Associate Product Manager at HashedIn by Deloitte, May 2024 - Present · 2 yrs 6 mos, Bengaluru · Hybrid, • Shipped a Retrieval-Augmented Generation insights feature on the Claude API into an enterprise B2B SaaS module — defined the model evaluation framework across 3 prompt variants with client stakeholders, cutting hallucination-related defects from ~12% to ~3% pre-launch and clearing go-live for a product line contributing $1.2M ARR. • Set the product roadmap and end-to-end delivery for that module across 4 quarters — prioritized requirements via RICE with a 7-member cross-functional partner team (engineering, design, QA), authored PRDs and detailed use cases for a 5-engineer squad across 18 sprints, and launched 6 releases while moving on-time delivery from 72% to 90% and cycle time down ~25%. • Analyzed funnel and cohort data in Amplitude using custom SQL to isolate 2 activation drop-off points, then redesigned onboarding through 10+ requirement-gathering workshops and structured UAT — lifting feature adoption from 18% to 31%, activation ~12% across 2 enterprise accounts, and time-to-value ~30%.
Software Development Engineer at HashedIn by Deloitte, May 2022 - May 2024 · 2 yrs 1 mo, Bengaluru · Hybrid, • Migrated a legacy multi-page interface to a single-page Angular application serving 10,000+ daily active users — reduced average page load time ~40% and raised Lighthouse performance scores across client-facing modules. • Built reusable Angular and React component libraries in TypeScript adopted by 3 project teams, accelerating frontend development on new features by ~20% and standardizing UI delivery across concurrent client engagements. • Raised unit-test coverage from ~35% to ~70% with Jasmine/Karma while implementing REST API integrations and reactive forms (RxJS) — reducing regression defects reaching QA and shortening release verification cycles.
Bachelor of Engineering - BE, Computer Science, Rungta College of Engineering & Technology Kohka-Kurud Bhilai
reema-sharma-ai-pm.vercel.app: https://reema-sharma-ai-pm.vercel.app/